@huggingface/inference 2.5.2 → 2.6.0

package/src/tasks/cv/imageToText.ts

@@ -20,7 +20,12 @@ export interface ImageToTextOutput {
  * This task reads some image input and outputs the text caption.
  */
 export async function imageToText(args: ImageToTextArgs, options?: Options): Promise<ImageToTextOutput> {
-	const res = (await request<[ImageToTextOutput]>(args, options))?.[0];
+	const res = (
+		await request<[ImageToTextOutput]>(args, {
+			...options,
+			taskHint: "image-to-text",
+		})
+	)?.[0];
 
 	if (typeof res?.generated_text !== "string") {
 		throw new InferenceOutputError("Expected {generated_text: string}");

package/src/tasks/cv/objectDetection.ts

@@ -37,7 +37,10 @@ export type ObjectDetectionOutput = ObjectDetectionOutputValue[];
  * Recommended model: facebook/detr-resnet-50
  */
 export async function objectDetection(args: ObjectDetectionArgs, options?: Options): Promise<ObjectDetectionOutput> {
-	const res = await request<ObjectDetectionOutput>(args, options);
+	const res = await request<ObjectDetectionOutput>(args, {
+		...options,
+		taskHint: "object-detection",
+	});
 	const isValidOutput =
 		Array.isArray(res) &&
 		res.every(

package/src/tasks/cv/textToImage.ts

@@ -39,7 +39,10 @@ export type TextToImageOutput = Blob;
  * Recommended model: stabilityai/stable-diffusion-2
  */
 export async function textToImage(args: TextToImageArgs, options?: Options): Promise<TextToImageOutput> {
-	const res = await request<TextToImageOutput>(args, options);
+	const res = await request<TextToImageOutput>(args, {
+		...options,
+		taskHint: "text-to-image",
+	});
 	const isValidOutput = res && res instanceof Blob;
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected Blob");

package/src/tasks/cv/zeroShotImageClassification.ts

@@ -45,7 +45,10 @@ export async function zeroShotImageClassification(
 		},
 	} as RequestArgs;
 
-	const res = await request<ZeroShotImageClassificationOutput>(reqArgs, options);
+	const res = await request<ZeroShotImageClassificationOutput>(reqArgs, {
+		...options,
+		taskHint: "zero-shot-image-classification",
+	});
 	const isValidOutput =
 		Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
 	if (!isValidOutput) {

package/src/tasks/multimodal/documentQuestionAnswering.ts

@@ -56,7 +56,10 @@ export async function documentQuestionAnswering(
 		},
 	} as RequestArgs;
 	const res = toArray(
-		await request<[DocumentQuestionAnsweringOutput] | DocumentQuestionAnsweringOutput>(reqArgs, options)
+		await request<[DocumentQuestionAnsweringOutput] | DocumentQuestionAnsweringOutput>(reqArgs, {
+			...options,
+			taskHint: "document-question-answering",
+		})
 	)?.[0];
 	const isValidOutput =
 		typeof res?.answer === "string" &&

package/src/tasks/multimodal/visualQuestionAnswering.ts

@@ -45,7 +45,12 @@ export async function visualQuestionAnswering(
 			),
 		},
 	} as RequestArgs;
-	const res = (await request<[VisualQuestionAnsweringOutput]>(reqArgs, options))?.[0];
+	const res = (
+		await request<[VisualQuestionAnsweringOutput]>(reqArgs, {
+			...options,
+			taskHint: "visual-question-answering",
+		})
+	)?.[0];
 	const isValidOutput = typeof res?.answer === "string" && typeof res.score === "number";
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected Array<{answer: string, score: number}>");

package/src/tasks/nlp/conversational.ts

@@ -63,7 +63,7 @@ export interface ConversationalOutput {
  *
  */
 export async function conversational(args: ConversationalArgs, options?: Options): Promise<ConversationalOutput> {
-	const res = await request<ConversationalOutput>(args, options);
+	const res = await request<ConversationalOutput>(args, { ...options, taskHint: "conversational" });
 	const isValidOutput =
 		Array.isArray(res.conversation.generated_responses) &&
 		res.conversation.generated_responses.every((x) => typeof x === "string") &&

package/src/tasks/nlp/featureExtraction.ts

@@ -25,16 +25,13 @@ export async function featureExtraction(
 	args: FeatureExtractionArgs,
 	options?: Options
 ): Promise<FeatureExtractionOutput> {
-	const defaultTask = await getDefaultTask(args.model, args.accessToken);
-	const res = await request<FeatureExtractionOutput>(
-		args,
-		defaultTask === "sentence-similarity"
-			? {
-					...options,
-					task: "feature-extraction",
-			  }
-			: options
-	);
+	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : undefined;
+
+	const res = await request<FeatureExtractionOutput>(args, {
+		...options,
+		taskHint: "feature-extraction",
+		...(defaultTask === "sentence-similarity" && { forceTask: "feature-extraction" }),
+	});
 	let isValidOutput = true;
 
 	const isNumArrayRec = (arr: unknown[], maxDepth: number, curDepth = 0): boolean => {

package/src/tasks/nlp/fillMask.ts

@@ -29,7 +29,10 @@ export type FillMaskOutput = {
  * Tries to fill in a hole with a missing word (token to be precise). That’s the base task for BERT models.
  */
 export async function fillMask(args: FillMaskArgs, options?: Options): Promise<FillMaskOutput> {
-	const res = await request<FillMaskOutput>(args, options);
+	const res = await request<FillMaskOutput>(args, {
+		...options,
+		taskHint: "fill-mask",
+	});
 	const isValidOutput =
 		Array.isArray(res) &&
 		res.every(

package/src/tasks/nlp/questionAnswering.ts

@@ -35,7 +35,10 @@ export async function questionAnswering(
 	args: QuestionAnsweringArgs,
 	options?: Options
 ): Promise<QuestionAnsweringOutput> {
-	const res = await request<QuestionAnsweringOutput>(args, options);
+	const res = await request<QuestionAnsweringOutput>(args, {
+		...options,
+		taskHint: "question-answering",
+	});
 	const isValidOutput =
 		typeof res === "object" &&
 		!!res &&

package/src/tasks/nlp/sentenceSimilarity.ts

@@ -25,16 +25,12 @@ export async function sentenceSimilarity(
 	args: SentenceSimilarityArgs,
 	options?: Options
 ): Promise<SentenceSimilarityOutput> {
-	const defaultTask = await getDefaultTask(args.model, args.accessToken);
-	const res = await request<SentenceSimilarityOutput>(
-		args,
-		defaultTask === "feature-extraction"
-			? {
-					...options,
-					task: "sentence-similarity",
-			  }
-			: options
-	);
+	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : undefined;
+	const res = await request<SentenceSimilarityOutput>(args, {
+		...options,
+		taskHint: "sentence-similarity",
+		...(defaultTask === "feature-extraction" && { forceTask: "sentence-similarity" }),
+	});
 
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
 	if (!isValidOutput) {

package/src/tasks/nlp/summarization.ts

@@ -50,7 +50,10 @@ export interface SummarizationOutput {
  * This task is well known to summarize longer text into shorter text. Be careful, some models have a maximum length of input. That means that the summary cannot handle full books for instance. Be careful when choosing your model.
  */
 export async function summarization(args: SummarizationArgs, options?: Options): Promise<SummarizationOutput> {
-	const res = await request<SummarizationOutput[]>(args, options);
+	const res = await request<SummarizationOutput[]>(args, {
+		...options,
+		taskHint: "summarization",
+	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.summary_text === "string");
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected Array<{summary_text: string}>");

package/src/tasks/nlp/tableQuestionAnswering.ts

@@ -41,7 +41,10 @@ export async function tableQuestionAnswering(
 	args: TableQuestionAnsweringArgs,
 	options?: Options
 ): Promise<TableQuestionAnsweringOutput> {
-	const res = await request<TableQuestionAnsweringOutput>(args, options);
+	const res = await request<TableQuestionAnsweringOutput>(args, {
+		...options,
+		taskHint: "table-question-answering",
+	});
 	const isValidOutput =
 		typeof res?.aggregator === "string" &&
 		typeof res.answer === "string" &&

package/src/tasks/nlp/textClassification.ts

@@ -27,7 +27,12 @@ export async function textClassification(
 	args: TextClassificationArgs,
 	options?: Options
 ): Promise<TextClassificationOutput> {
-	const res = (await request<TextClassificationOutput[]>(args, options))?.[0];
+	const res = (
+		await request<TextClassificationOutput[]>(args, {
+			...options,
+			taskHint: "text-classification",
+		})
+	)?.[0];
 	const isValidOutput =
 		Array.isArray(res) && res.every((x) => typeof x?.label === "string" && typeof x.score === "number");
 	if (!isValidOutput) {

package/src/tasks/nlp/textGeneration.ts

@@ -62,7 +62,10 @@ export interface TextGenerationOutput {
  * Use to continue text from a prompt. This is a very generic task. Recommended model: gpt2 (it’s a simple model, but fun to play with).
  */
 export async function textGeneration(args: TextGenerationArgs, options?: Options): Promise<TextGenerationOutput> {
-	const res = await request<TextGenerationOutput[]>(args, options);
+	const res = await request<TextGenerationOutput[]>(args, {
+		...options,
+		taskHint: "text-generation",
+	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.generated_text === "string");
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected Array<{generated_text: string}>");

package/src/tasks/nlp/textGenerationStream.ts

@@ -88,5 +88,8 @@ export async function* textGenerationStream(
 	args: TextGenerationArgs,
 	options?: Options
 ): AsyncGenerator<TextGenerationStreamOutput> {
-	yield* streamingRequest<TextGenerationStreamOutput>(args, options);
+	yield* streamingRequest<TextGenerationStreamOutput>(args, {
+		...options,
+		taskHint: "text-generation",
+	});
 }

package/src/tasks/nlp/tokenClassification.ts

@@ -58,7 +58,12 @@ export async function tokenClassification(
 	args: TokenClassificationArgs,
 	options?: Options
 ): Promise<TokenClassificationOutput> {
-	const res = toArray(await request<TokenClassificationOutput[number] | TokenClassificationOutput>(args, options));
+	const res = toArray(
+		await request<TokenClassificationOutput[number] | TokenClassificationOutput>(args, {
+			...options,
+			taskHint: "token-classification",
+		})
+	);
 	const isValidOutput =
 		Array.isArray(res) &&
 		res.every(

package/src/tasks/nlp/translation.ts

@@ -20,7 +20,10 @@ export interface TranslationOutput {
  * This task is well known to translate text from one language to another. Recommended model: Helsinki-NLP/opus-mt-ru-en.
  */
 export async function translation(args: TranslationArgs, options?: Options): Promise<TranslationOutput> {
-	const res = await request<TranslationOutput[]>(args, options);
+	const res = await request<TranslationOutput[]>(args, {
+		...options,
+		taskHint: "translation",
+	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.translation_text === "string");
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected type Array<{translation_text: string}>");

package/src/tasks/nlp/zeroShotClassification.ts

@@ -36,7 +36,10 @@ export async function zeroShotClassification(
 	options?: Options
 ): Promise<ZeroShotClassificationOutput> {
 	const res = toArray(
-		await request<ZeroShotClassificationOutput[number] | ZeroShotClassificationOutput>(args, options)
+		await request<ZeroShotClassificationOutput[number] | ZeroShotClassificationOutput>(args, {
+			...options,
+			taskHint: "zero-shot-classification",
+		})
 	);
 	const isValidOutput =
 		Array.isArray(res) &&

package/src/tasks/tabular/tabularClassification.ts

@@ -25,7 +25,10 @@ export async function tabularClassification(
 	args: TabularClassificationArgs,
 	options?: Options
 ): Promise<TabularClassificationOutput> {
-	const res = await request<TabularClassificationOutput>(args, options);
+	const res = await request<TabularClassificationOutput>(args, {
+		...options,
+		taskHint: "tabular-classification",
+	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected number[]");

package/src/tasks/tabular/tabularRegression.ts

@@ -25,7 +25,10 @@ export async function tabularRegression(
 	args: TabularRegressionArgs,
 	options?: Options
 ): Promise<TabularRegressionOutput> {
-	const res = await request<TabularRegressionOutput>(args, options);
+	const res = await request<TabularRegressionOutput>(args, {
+		...options,
+		taskHint: "tabular-regression",
+	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected number[]");

package/src/types.ts

@@ -26,7 +26,39 @@ export interface Options {
 	fetch?: typeof fetch;
 }
 
-export type InferenceTask = "text-classification" | "feature-extraction" | "sentence-similarity";
+export type InferenceTask =
+	| "audio-classification"
+	| "audio-to-audio"
+	| "automatic-speech-recognition"
+	| "conversational"
+	| "depth-estimation"
+	| "document-question-answering"
+	| "feature-extraction"
+	| "fill-mask"
+	| "image-classification"
+	| "image-segmentation"
+	| "image-to-image"
+	| "image-to-text"
+	| "object-detection"
+	| "video-classification"
+	| "question-answering"
+	| "reinforcement-learning"
+	| "sentence-similarity"
+	| "summarization"
+	| "table-question-answering"
+	| "tabular-classification"
+	| "tabular-regression"
+	| "text-classification"
+	| "text-generation"
+	| "text-to-image"
+	| "text-to-speech"
+	| "text-to-video"
+	| "token-classification"
+	| "translation"
+	| "unconditional-image-generation"
+	| "visual-question-answering"
+	| "zero-shot-classification"
+	| "zero-shot-image-classification";
 
 export interface BaseArgs {
 	/**
@@ -37,8 +69,10 @@ export interface BaseArgs {
 	accessToken?: string;
 	/**
 	 * The model to use. Can be a full URL for HF inference endpoints.
+	 *
+	 * If not specified, will call huggingface.co/api/tasks to get the default model for the task.
 	 */
-	model: string;
+	model?: string;
 }
 
 export type RequestArgs = BaseArgs &