[inference] Proposal: rename taskHint param to task (#1204)

julien-c · web-flow · commit befb723fe4c8 · 2025-02-24T15:01:23.000+01:00
so we merge `task` and `taskHint`
diff --git a/packages/inference/src/lib/getProviderModelId.ts b/packages/inference/src/lib/getProviderModelId.ts
@@ -15,19 +15,19 @@ export async function getProviderModelId(
 	},
 	args: RequestArgs,
 	options: {
-		taskHint?: InferenceTask;
+		task?: InferenceTask;
 		chatCompletion?: boolean;
 		fetch?: Options["fetch"];
 	} = {}
 ): Promise<string> {
 	if (params.provider === "hf-inference") {
 		return params.model;
 	}
-	if (!options.taskHint) {
-		throw new Error("taskHint must be specified when using a third-party provider");
+	if (!options.task) {
+		throw new Error("task must be specified when using a third-party provider");
 	}
 	const task: WidgetType =
-		options.taskHint === "text-generation" && options.chatCompletion ? "conversational" : options.taskHint;
+		options.task === "text-generation" && options.chatCompletion ? "conversational" : options.task;
 
 	// A dict called HARDCODED_MODEL_ID_MAPPING takes precedence in all cases (useful for dev purposes)
 	if (HARDCODED_MODEL_ID_MAPPING[params.provider]?.[params.model]) {
diff --git a/packages/inference/src/lib/makeRequestOptions.ts b/packages/inference/src/lib/makeRequestOptions.ts
@@ -31,30 +31,30 @@ export async function makeRequestOptions(
 		stream?: boolean;
 	},
 	options?: Options & {
-		/** To load default model if needed */
-		taskHint?: InferenceTask;
+		/** In most cases (unless we pass a endpointUrl) we know the task */
+		task?: InferenceTask;
 		chatCompletion?: boolean;
 	}
 ): Promise<{ url: string; info: RequestInit }> {
 	const { accessToken, endpointUrl, provider: maybeProvider, model: maybeModel, ...remainingArgs } = args;
 	let otherArgs = remainingArgs;
 	const provider = maybeProvider ?? "hf-inference";
 
-	const { includeCredentials, taskHint, chatCompletion } = options ?? {};
+	const { includeCredentials, task, chatCompletion } = options ?? {};
 
 	if (endpointUrl && provider !== "hf-inference") {
 		throw new Error(`Cannot use endpointUrl with a third-party provider.`);
 	}
 	if (maybeModel && isUrl(maybeModel)) {
 		throw new Error(`Model URLs are no longer supported. Use endpointUrl instead.`);
 	}
-	if (!maybeModel && !taskHint) {
+	if (!maybeModel && !task) {
 		throw new Error("No model provided, and no task has been specified.");
 	}
 	// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
-	const hfModel = maybeModel ?? (await loadDefaultModel(taskHint!));
+	const hfModel = maybeModel ?? (await loadDefaultModel(task!));
 	const model = await getProviderModelId({ model: hfModel, provider }, args, {
-		taskHint,
+		task,
 		chatCompletion,
 		fetch: options?.fetch,
 	});
@@ -77,7 +77,7 @@ export async function makeRequestOptions(
 				chatCompletion: chatCompletion ?? false,
 				model,
 				provider: provider ?? "hf-inference",
-				taskHint,
+				task,
 		  });
 
 	const headers: Record<string, string> = {};
@@ -133,7 +133,7 @@ export async function makeRequestOptions(
 			? args.data
 			: JSON.stringify({
 					...otherArgs,
-					...(taskHint === "text-to-image" && provider === "hyperbolic"
+					...(task === "text-to-image" && provider === "hyperbolic"
 						? { model_name: model }
 						: chatCompletion || provider === "together" || provider === "nebius" || provider === "hyperbolic"
 						  ? { model }
@@ -151,7 +151,7 @@ function makeUrl(params: {
 	chatCompletion: boolean;
 	model: string;
 	provider: InferenceProvider;
-	taskHint: InferenceTask | undefined;
+	task: InferenceTask | undefined;
 }): string {
 	if (params.authMethod === "none" && params.provider !== "hf-inference") {
 		throw new Error("Authentication is required when requesting a third-party provider. Please provide accessToken");
@@ -176,10 +176,10 @@ function makeUrl(params: {
 				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
 				: NEBIUS_API_BASE_URL;
 
-			if (params.taskHint === "text-to-image") {
+			if (params.task === "text-to-image") {
 				return `${baseUrl}/v1/images/generations`;
 			}
-			if (params.taskHint === "text-generation") {
+			if (params.task === "text-generation") {
 				if (params.chatCompletion) {
 					return `${baseUrl}/v1/chat/completions`;
 				}
@@ -203,7 +203,7 @@ function makeUrl(params: {
 				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
 				: SAMBANOVA_API_BASE_URL;
 			/// Sambanova API matches OpenAI-like APIs: model is defined in the request body
-			if (params.taskHint === "text-generation" && params.chatCompletion) {
+			if (params.task === "text-generation" && params.chatCompletion) {
 				return `${baseUrl}/v1/chat/completions`;
 			}
 			return baseUrl;
@@ -213,10 +213,10 @@ function makeUrl(params: {
 				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
 				: TOGETHER_API_BASE_URL;
 			/// Together API matches OpenAI-like APIs: model is defined in the request body
-			if (params.taskHint === "text-to-image") {
+			if (params.task === "text-to-image") {
 				return `${baseUrl}/v1/images/generations`;
 			}
-			if (params.taskHint === "text-generation") {
+			if (params.task === "text-generation") {
 				if (params.chatCompletion) {
 					return `${baseUrl}/v1/chat/completions`;
 				}
@@ -229,7 +229,7 @@ function makeUrl(params: {
 			const baseUrl = shouldProxy
 				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
 				: FIREWORKS_AI_API_BASE_URL;
-			if (params.taskHint === "text-generation" && params.chatCompletion) {
+			if (params.task === "text-generation" && params.chatCompletion) {
 				return `${baseUrl}/v1/chat/completions`;
 			}
 			return baseUrl;
@@ -239,7 +239,7 @@ function makeUrl(params: {
 				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
 				: HYPERBOLIC_API_BASE_URL;
 
-			if (params.taskHint === "text-to-image") {
+			if (params.task === "text-to-image") {
 				return `${baseUrl}/v1/images/generations`;
 			}
 			return `${baseUrl}/v1/chat/completions`;
@@ -248,7 +248,7 @@ function makeUrl(params: {
 			const baseUrl = shouldProxy
 				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
 				: NOVITA_API_BASE_URL;
-			if (params.taskHint === "text-generation") {
+			if (params.task === "text-generation") {
 				if (params.chatCompletion) {
 					return `${baseUrl}/chat/completions`;
 				}
@@ -258,11 +258,11 @@ function makeUrl(params: {
 		}
 		default: {
 			const baseUrl = HF_HUB_INFERENCE_PROXY_TEMPLATE.replaceAll("{{PROVIDER}}", "hf-inference");
-			if (params.taskHint && ["feature-extraction", "sentence-similarity"].includes(params.taskHint)) {
+			if (params.task && ["feature-extraction", "sentence-similarity"].includes(params.task)) {
 				/// when deployed on hf-inference, those two tasks are automatically compatible with one another.
-				return `${baseUrl}/pipeline/${params.taskHint}/${params.model}`;
+				return `${baseUrl}/pipeline/${params.task}/${params.model}`;
 			}
-			if (params.taskHint === "text-generation" && params.chatCompletion) {
+			if (params.task === "text-generation" && params.chatCompletion) {
 				return `${baseUrl}/models/${params.model}/v1/chat/completions`;
 			}
 			return `${baseUrl}/models/${params.model}`;
diff --git a/packages/inference/src/tasks/audio/audioClassification.ts b/packages/inference/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");
diff --git a/packages/inference/src/tasks/audio/audioToAudio.ts b/packages/inference/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);
diff --git a/packages/inference/src/tasks/audio/automaticSpeechRecognition.ts b/packages/inference/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) {
diff --git a/packages/inference/src/tasks/audio/textToSpeech.ts b/packages/inference/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;
diff --git a/packages/inference/src/tasks/custom/request.ts b/packages/inference/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;
 	}
diff --git a/packages/inference/src/tasks/custom/streamingRequest.ts b/packages/inference/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;
 	}
diff --git a/packages/inference/src/tasks/cv/imageClassification.ts b/packages/inference/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");
diff --git a/packages/inference/src/tasks/cv/imageSegmentation.ts b/packages/inference/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) &&
diff --git a/packages/inference/src/tasks/cv/imageToImage.ts b/packages/inference/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) {
diff --git a/packages/inference/src/tasks/cv/imageToText.ts b/packages/inference/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];
 
diff --git a/packages/inference/src/tasks/cv/objectDetection.ts b/packages/inference/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) &&
diff --git a/packages/inference/src/tasks/cv/textToImage.ts b/packages/inference/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") {
diff --git a/packages/inference/src/tasks/cv/textToVideo.ts b/packages/inference/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") {
diff --git a/packages/inference/src/tasks/cv/zeroShotImageClassification.ts b/packages/inference/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");
diff --git a/packages/inference/src/tasks/multimodal/documentQuestionAnswering.ts b/packages/inference/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",
 		})
 	);
 
diff --git a/packages/inference/src/tasks/multimodal/visualQuestionAnswering.ts b/packages/inference/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) &&
diff --git a/packages/inference/src/tasks/nlp/chatCompletion.ts b/packages/inference/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,
 	});
 
diff --git a/packages/inference/src/tasks/nlp/chatCompletionStream.ts b/packages/inference/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,
 	});
 }
diff --git a/packages/inference/src/tasks/nlp/featureExtraction.ts b/packages/inference/src/tasks/nlp/featureExtraction.ts
@@ -26,7 +26,7 @@ export async function featureExtraction(
 ): Promise<FeatureExtractionOutput> {
 	const res = await request<FeatureExtractionOutput>(args, {
 		...options,
-		taskHint: "feature-extraction",
+		task: "feature-extraction",
 	});
 	let isValidOutput = true;
 
diff --git a/packages/inference/src/tasks/nlp/fillMask.ts b/packages/inference/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) &&
diff --git a/packages/inference/src/tasks/nlp/questionAnswering.ts b/packages/inference/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(
diff --git a/packages/inference/src/tasks/nlp/sentenceSimilarity.ts b/packages/inference/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");
diff --git a/packages/inference/src/tasks/nlp/summarization.ts b/packages/inference/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) {
diff --git a/packages/inference/src/tasks/nlp/tableQuestionAnswering.ts b/packages/inference/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) {
diff --git a/packages/inference/src/tasks/nlp/textClassification.ts b/packages/inference/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 =
diff --git a/packages/inference/src/tasks/nlp/textGeneration.ts b/packages/inference/src/tasks/nlp/textGeneration.ts
diff --git a/packages/inference/src/tasks/nlp/textGenerationStream.ts b/packages/inference/src/tasks/nlp/textGenerationStream.ts
diff --git a/packages/inference/src/tasks/nlp/tokenClassification.ts b/packages/inference/src/tasks/nlp/tokenClassification.ts
diff --git a/packages/inference/src/tasks/nlp/translation.ts b/packages/inference/src/tasks/nlp/translation.ts
diff --git a/packages/inference/src/tasks/nlp/zeroShotClassification.ts b/packages/inference/src/tasks/nlp/zeroShotClassification.ts
diff --git a/packages/inference/src/tasks/tabular/tabularClassification.ts b/packages/inference/src/tasks/tabular/tabularClassification.ts
diff --git a/packages/inference/src/tasks/tabular/tabularRegression.ts b/packages/inference/src/tasks/tabular/tabularRegression.ts
