@huggingface/inference 3.3.6 → 3.3.7

package/src/providers/together.ts

@@ -1,5 +1,3 @@
-export const TOGETHER_API_BASE_URL = "https://api.together.xyz";
-
 /**
  * See the registered mapping of HF model ID => Together model ID here:
  *
@@ -16,3 +14,37 @@ export const TOGETHER_API_BASE_URL = "https://api.together.xyz";
  *
  * Thanks!
  */
+import type { ProviderConfig, UrlParams, HeaderParams, BodyParams } from "../types";
+
+const TOGETHER_API_BASE_URL = "https://api.together.xyz";
+
+const makeBody = (params: BodyParams): Record<string, unknown> => {
+	return {
+		...params.args,
+		model: params.model,
+	};
+};
+
+const makeHeaders = (params: HeaderParams): Record<string, string> => {
+	return { Authorization: `Bearer ${params.accessToken}` };
+};
+
+const makeUrl = (params: UrlParams): string => {
+	if (params.task === "text-to-image") {
+		return `${params.baseUrl}/v1/images/generations`;
+	}
+	if (params.task === "text-generation") {
+		if (params.chatCompletion) {
+			return `${params.baseUrl}/v1/chat/completions`;
+		}
+		return `${params.baseUrl}/v1/completions`;
+	}
+	return params.baseUrl;
+};
+
+export const TOGETHER_CONFIG: ProviderConfig = {
+	baseUrl: TOGETHER_API_BASE_URL,
+	makeBody,
+	makeHeaders,
+	makeUrl,
+};

package/src/tasks/audio/audioClassification.ts

@@ -18,7 +18,7 @@ export async function audioClassification(
 	const payload = preparePayload(args);
 	const res = await request<AudioClassificationOutput>(payload, {
 		...options,
-		taskHint: "audio-classification",
+		task: "audio-classification",
 	});
 	const isValidOutput =
 		Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");

package/src/tasks/audio/audioToAudio.ts

@@ -39,7 +39,7 @@ export async function audioToAudio(args: AudioToAudioArgs, options?: Options): P
 	const payload = preparePayload(args);
 	const res = await request<AudioToAudioOutput>(payload, {
 		...options,
-		taskHint: "audio-to-audio",
+		task: "audio-to-audio",
 	});
 
 	return validateOutput(res);

package/src/tasks/audio/automaticSpeechRecognition.ts

@@ -19,7 +19,7 @@ export async function automaticSpeechRecognition(
 	const payload = await buildPayload(args);
 	const res = await request<AutomaticSpeechRecognitionOutput>(payload, {
 		...options,
-		taskHint: "automatic-speech-recognition",
+		task: "automatic-speech-recognition",
 	});
 	const isValidOutput = typeof res?.text === "string";
 	if (!isValidOutput) {

package/src/tasks/audio/textToSpeech.ts

@@ -24,7 +24,7 @@ export async function textToSpeech(args: TextToSpeechArgs, options?: Options): P
 			: args;
 	const res = await request<Blob | OutputUrlTextToSpeechGeneration>(payload, {
 		...options,
-		taskHint: "text-to-speech",
+		task: "text-to-speech",
 	});
 	if (res instanceof Blob) {
 		return res;

package/src/tasks/custom/request.ts

@@ -7,10 +7,8 @@ import { makeRequestOptions } from "../../lib/makeRequestOptions";
 export async function request<T>(
 	args: RequestArgs,
 	options?: Options & {
-		/** When a model can be used for multiple tasks, and we want to run a non-default task */
-		task?: string | InferenceTask;
-		/** To load default model if needed */
-		taskHint?: InferenceTask;
+		/** In most cases (unless we pass a endpointUrl) we know the task */
+		task?: InferenceTask;
 		/** Is chat completion compatible */
 		chatCompletion?: boolean;
 	}

package/src/tasks/custom/streamingRequest.ts

@@ -9,10 +9,8 @@ import { getLines, getMessages } from "../../vendor/fetch-event-source/parse";
 export async function* streamingRequest<T>(
 	args: RequestArgs,
 	options?: Options & {
-		/** When a model can be used for multiple tasks, and we want to run a non-default task */
-		task?: string | InferenceTask;
-		/** To load default model if needed */
-		taskHint?: InferenceTask;
+		/** In most cases (unless we pass a endpointUrl) we know the task */
+		task?: InferenceTask;
 		/** Is chat completion compatible */
 		chatCompletion?: boolean;
 	}

package/src/tasks/cv/imageClassification.ts

@@ -17,7 +17,7 @@ export async function imageClassification(
 	const payload = preparePayload(args);
 	const res = await request<ImageClassificationOutput>(payload, {
 		...options,
-		taskHint: "image-classification",
+		task: "image-classification",
 	});
 	const isValidOutput =
 		Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");

package/src/tasks/cv/imageSegmentation.ts

@@ -17,7 +17,7 @@ export async function imageSegmentation(
 	const payload = preparePayload(args);
 	const res = await request<ImageSegmentationOutput>(payload, {
 		...options,
-		taskHint: "image-segmentation",
+		task: "image-segmentation",
 	});
 	const isValidOutput =
 		Array.isArray(res) &&

package/src/tasks/cv/imageToImage.ts

@@ -28,7 +28,7 @@ export async function imageToImage(args: ImageToImageArgs, options?: Options): P
 	}
 	const res = await request<Blob>(reqArgs, {
 		...options,
-		taskHint: "image-to-image",
+		task: "image-to-image",
 	});
 	const isValidOutput = res && res instanceof Blob;
 	if (!isValidOutput) {

package/src/tasks/cv/imageToText.ts

@@ -14,7 +14,7 @@ export async function imageToText(args: ImageToTextArgs, options?: Options): Pro
 	const res = (
 		await request<[ImageToTextOutput]>(payload, {
 			...options,
-			taskHint: "image-to-text",
+			task: "image-to-text",
 		})
 	)?.[0];
 

package/src/tasks/cv/objectDetection.ts

@@ -14,7 +14,7 @@ export async function objectDetection(args: ObjectDetectionArgs, options?: Optio
 	const payload = preparePayload(args);
 	const res = await request<ObjectDetectionOutput>(payload, {
 		...options,
-		taskHint: "object-detection",
+		task: "object-detection",
 	});
 	const isValidOutput =
 		Array.isArray(res) &&

package/src/tasks/cv/textToImage.ts

@@ -73,7 +73,7 @@ export async function textToImage(args: TextToImageArgs, options?: TextToImageOp
 		| HyperbolicTextToImageOutput
 	>(payload, {
 		...options,
-		taskHint: "text-to-image",
+		task: "text-to-image",
 	});
 
 	if (res && typeof res === "object") {

package/src/tasks/cv/textToVideo.ts

@@ -35,7 +35,7 @@ export async function textToVideo(args: TextToVideoArgs, options?: Options): Pro
 			: args;
 	const res = await request<FalAiOutput | ReplicateOutput>(payload, {
 		...options,
-		taskHint: "text-to-video",
+		task: "text-to-video",
 	});
 
 	if (args.provider === "fal-ai") {

package/src/tasks/cv/zeroShotImageClassification.ts

@@ -48,7 +48,7 @@ export async function zeroShotImageClassification(
 	const payload = await preparePayload(args);
 	const res = await request<ZeroShotImageClassificationOutput>(payload, {
 		...options,
-		taskHint: "zero-shot-image-classification",
+		task: "zero-shot-image-classification",
 	});
 	const isValidOutput =
 		Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");

package/src/tasks/multimodal/documentQuestionAnswering.ts

@@ -32,7 +32,7 @@ export async function documentQuestionAnswering(
 	const res = toArray(
 		await request<DocumentQuestionAnsweringOutput | DocumentQuestionAnsweringOutput[number]>(reqArgs, {
 			...options,
-			taskHint: "document-question-answering",
+			task: "document-question-answering",
 		})
 	);
 

package/src/tasks/multimodal/visualQuestionAnswering.ts

@@ -29,7 +29,7 @@ export async function visualQuestionAnswering(
 	} as RequestArgs;
 	const res = await request<VisualQuestionAnsweringOutput>(reqArgs, {
 		...options,
-		taskHint: "visual-question-answering",
+		task: "visual-question-answering",
 	});
 	const isValidOutput =
 		Array.isArray(res) &&

package/src/tasks/nlp/chatCompletion.ts

@@ -12,7 +12,7 @@ export async function chatCompletion(
 ): Promise<ChatCompletionOutput> {
 	const res = await request<ChatCompletionOutput>(args, {
 		...options,
-		taskHint: "text-generation",
+		task: "text-generation",
 		chatCompletion: true,
 	});
 

package/src/tasks/nlp/chatCompletionStream.ts

@@ -11,7 +11,7 @@ export async function* chatCompletionStream(
 ): AsyncGenerator<ChatCompletionStreamOutput> {
 	yield* streamingRequest<ChatCompletionStreamOutput>(args, {
 		...options,
-		taskHint: "text-generation",
+		task: "text-generation",
 		chatCompletion: true,
 	});
 }

package/src/tasks/nlp/featureExtraction.ts

@@ -1,16 +1,9 @@
+import type { FeatureExtractionInput } from "@huggingface/tasks";
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import type { BaseArgs, Options } from "../../types";
 import { request } from "../custom/request";
 
-export type FeatureExtractionArgs = BaseArgs & {
-	/**
-	 *  The inputs is a string or a list of strings to get the features from.
-	 *
-	 *  inputs: "That is a happy person",
-	 *
-	 */
-	inputs: string | string[];
-};
+export type FeatureExtractionArgs = BaseArgs & FeatureExtractionInput;
 
 /**
  * Returned values are a multidimensional array of floats (dimension depending on if you sent a string or a list of string, and if the automatic reduction, usually mean_pooling for instance was applied for you or not. This should be explained on the model's README).
@@ -26,7 +19,7 @@ export async function featureExtraction(
 ): Promise<FeatureExtractionOutput> {
 	const res = await request<FeatureExtractionOutput>(args, {
 		...options,
-		taskHint: "feature-extraction",
+		task: "feature-extraction",
 	});
 	let isValidOutput = true;
 

package/src/tasks/nlp/fillMask.ts

@@ -11,7 +11,7 @@ export type FillMaskArgs = BaseArgs & FillMaskInput;
 export async function fillMask(args: FillMaskArgs, options?: Options): Promise<FillMaskOutput> {
 	const res = await request<FillMaskOutput>(args, {
 		...options,
-		taskHint: "fill-mask",
+		task: "fill-mask",
 	});
 	const isValidOutput =
 		Array.isArray(res) &&

package/src/tasks/nlp/questionAnswering.ts

@@ -14,7 +14,7 @@ export async function questionAnswering(
 ): Promise<QuestionAnsweringOutput[number]> {
 	const res = await request<QuestionAnsweringOutput | QuestionAnsweringOutput[number]>(args, {
 		...options,
-		taskHint: "question-answering",
+		task: "question-answering",
 	});
 	const isValidOutput = Array.isArray(res)
 		? res.every(

package/src/tasks/nlp/sentenceSimilarity.ts

@@ -15,7 +15,7 @@ export async function sentenceSimilarity(
 ): Promise<SentenceSimilarityOutput> {
 	const res = await request<SentenceSimilarityOutput>(prepareInput(args), {
 		...options,
-		taskHint: "sentence-similarity",
+		task: "sentence-similarity",
 	});
 
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");

package/src/tasks/nlp/summarization.ts

@@ -11,7 +11,7 @@ export type SummarizationArgs = BaseArgs & SummarizationInput;
 export async function summarization(args: SummarizationArgs, options?: Options): Promise<SummarizationOutput> {
 	const res = await request<SummarizationOutput[]>(args, {
 		...options,
-		taskHint: "summarization",
+		task: "summarization",
 	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.summary_text === "string");
 	if (!isValidOutput) {

package/src/tasks/nlp/tableQuestionAnswering.ts

@@ -14,7 +14,7 @@ export async function tableQuestionAnswering(
 ): Promise<TableQuestionAnsweringOutput[number]> {
 	const res = await request<TableQuestionAnsweringOutput | TableQuestionAnsweringOutput[number]>(args, {
 		...options,
-		taskHint: "table-question-answering",
+		task: "table-question-answering",
 	});
 	const isValidOutput = Array.isArray(res) ? res.every((elem) => validate(elem)) : validate(res);
 	if (!isValidOutput) {

package/src/tasks/nlp/textClassification.ts

@@ -15,7 +15,7 @@ export async function textClassification(
 	const res = (
 		await request<TextClassificationOutput>(args, {
 			...options,
-			taskHint: "text-classification",
+			task: "text-classification",
 		})
 	)?.[0];
 	const isValidOutput =

package/src/tasks/nlp/textGeneration.ts

@@ -39,7 +39,7 @@ export async function textGeneration(
 		args.prompt = args.inputs;
 		const raw = await request<TogeteherTextCompletionOutput>(args, {
 			...options,
-			taskHint: "text-generation",
+			task: "text-generation",
 		});
 		const isValidOutput =
 			typeof raw === "object" && "choices" in raw && Array.isArray(raw?.choices) && typeof raw?.model === "string";
@@ -63,7 +63,7 @@ export async function textGeneration(
 		};
 		const raw = await request<HyperbolicTextCompletionOutput>(payload, {
 			...options,
-			taskHint: "text-generation",
+			task: "text-generation",
 		});
 		const isValidOutput =
 			typeof raw === "object" && "choices" in raw && Array.isArray(raw?.choices) && typeof raw?.model === "string";
@@ -78,7 +78,7 @@ export async function textGeneration(
 		const res = toArray(
 			await request<TextGenerationOutput | TextGenerationOutput[]>(args, {
 				...options,
-				taskHint: "text-generation",
+				task: "text-generation",
 			})
 		);
 

package/src/tasks/nlp/textGenerationStream.ts

@@ -91,6 +91,6 @@ export async function* textGenerationStream(
 ): AsyncGenerator<TextGenerationStreamOutput> {
 	yield* streamingRequest<TextGenerationStreamOutput>(args, {
 		...options,
-		taskHint: "text-generation",
+		task: "text-generation",
 	});
 }

package/src/tasks/nlp/tokenClassification.ts

@@ -16,7 +16,7 @@ export async function tokenClassification(
 	const res = toArray(
 		await request<TokenClassificationOutput[number] | TokenClassificationOutput>(args, {
 			...options,
-			taskHint: "token-classification",
+			task: "token-classification",
 		})
 	);
 	const isValidOutput =

package/src/tasks/nlp/translation.ts

@@ -10,7 +10,7 @@ export type TranslationArgs = BaseArgs & TranslationInput;
 export async function translation(args: TranslationArgs, options?: Options): Promise<TranslationOutput> {
 	const res = await request<TranslationOutput>(args, {
 		...options,
-		taskHint: "translation",
+		task: "translation",
 	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x?.translation_text === "string");
 	if (!isValidOutput) {

package/src/tasks/nlp/zeroShotClassification.ts

@@ -16,7 +16,7 @@ export async function zeroShotClassification(
 	const res = toArray(
 		await request<ZeroShotClassificationOutput[number] | ZeroShotClassificationOutput>(args, {
 			...options,
-			taskHint: "zero-shot-classification",
+			task: "zero-shot-classification",
 		})
 	);
 	const isValidOutput =

package/src/tasks/tabular/tabularClassification.ts

@@ -27,7 +27,7 @@ export async function tabularClassification(
 ): Promise<TabularClassificationOutput> {
 	const res = await request<TabularClassificationOutput>(args, {
 		...options,
-		taskHint: "tabular-classification",
+		task: "tabular-classification",
 	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
 	if (!isValidOutput) {

package/src/tasks/tabular/tabularRegression.ts

@@ -27,7 +27,7 @@ export async function tabularRegression(
 ): Promise<TabularRegressionOutput> {
 	const res = await request<TabularRegressionOutput>(args, {
 		...options,
-		taskHint: "tabular-regression",
+		task: "tabular-regression",
 	});
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");
 	if (!isValidOutput) {

package/src/types.ts

@@ -1,4 +1,4 @@
-import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
+import type { ChatCompletionInput, FeatureExtractionInput, PipelineType } from "@huggingface/tasks";
 
 /**
  * HF model id, like "meta-llama/Llama-3.3-70B-Instruct"
@@ -86,7 +86,33 @@ export type RequestArgs = BaseArgs &
 		| { text: string }
 		| { audio_url: string }
 		| ChatCompletionInput
+		| FeatureExtractionInput
 	) & {
 		parameters?: Record<string, unknown>;
-		accessToken?: string;
 	};
+
+export interface ProviderConfig {
+	baseUrl: string;
+	makeBody: (params: BodyParams) => Record<string, unknown>;
+	makeHeaders: (params: HeaderParams) => Record<string, string>;
+	makeUrl: (params: UrlParams) => string;
+}
+
+export interface HeaderParams {
+	accessToken?: string;
+	authMethod: "none" | "hf-token" | "credentials-include" | "provider-key";
+}
+
+export interface UrlParams {
+	baseUrl: string;
+	model: string;
+	task?: InferenceTask;
+	chatCompletion?: boolean;
+}
+
+export interface BodyParams {
+	args: Record<string, unknown>;
+	chatCompletion?: boolean;
+	model: string;
+	task?: InferenceTask;
+}