@huggingface/inference 3.9.2 → 3.11.0

package/src/tasks/cv/objectDetection.ts

@@ -1,4 +1,5 @@
 import type { ObjectDetectionInput, ObjectDetectionOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -11,7 +12,8 @@ export type ObjectDetectionArgs = BaseArgs & (ObjectDetectionInput | LegacyImage
  * Recommended model: facebook/detr-resnet-50
  */
 export async function objectDetection(args: ObjectDetectionArgs, options?: Options): Promise<ObjectDetectionOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "object-detection");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "object-detection");
 	const payload = preparePayload(args);
 	const { data: res } = await innerRequest<ObjectDetectionOutput>(payload, providerHelper, {
 		...options,

package/src/tasks/cv/textToImage.ts

@@ -1,4 +1,5 @@
 import type { TextToImageInput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import { makeRequestOptions } from "../../lib/makeRequestOptions";
 import type { BaseArgs, Options } from "../../types";
@@ -23,7 +24,7 @@ export async function textToImage(
 	options?: TextToImageOptions & { outputType?: undefined | "blob" }
 ): Promise<Blob>;
 export async function textToImage(args: TextToImageArgs, options?: TextToImageOptions): Promise<Blob | string> {
-	const provider = args.provider ?? "hf-inference";
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
 	const providerHelper = getProviderHelper(provider, "text-to-image");
 	const { data: res } = await innerRequest<Record<string, unknown>>(args, providerHelper, {
 		...options,

package/src/tasks/cv/textToVideo.ts

@@ -1,4 +1,5 @@
 import type { TextToVideoInput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import { makeRequestOptions } from "../../lib/makeRequestOptions";
 import type { FalAiQueueOutput } from "../../providers/fal-ai";
@@ -12,7 +13,7 @@ export type TextToVideoArgs = BaseArgs & TextToVideoInput;
 export type TextToVideoOutput = Blob;
 
 export async function textToVideo(args: TextToVideoArgs, options?: Options): Promise<TextToVideoOutput> {
-	const provider = args.provider ?? "hf-inference";
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
 	const providerHelper = getProviderHelper(provider, "text-to-video");
 	const { data: response } = await innerRequest<FalAiQueueOutput | ReplicateOutput | NovitaOutput>(
 		args,

package/src/tasks/cv/zeroShotImageClassification.ts

@@ -1,4 +1,5 @@
 import type { ZeroShotImageClassificationInput, ZeroShotImageClassificationOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options, RequestArgs } from "../../types";
 import { base64FromBytes } from "../../utils/base64FromBytes";
@@ -44,7 +45,8 @@ export async function zeroShotImageClassification(
 	args: ZeroShotImageClassificationArgs,
 	options?: Options
 ): Promise<ZeroShotImageClassificationOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "zero-shot-image-classification");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "zero-shot-image-classification");
 	const payload = await preparePayload(args);
 	const { data: res } = await innerRequest<ZeroShotImageClassificationOutput>(payload, providerHelper, {
 		...options,

package/src/tasks/multimodal/documentQuestionAnswering.ts

@@ -3,6 +3,7 @@ import type {
 	DocumentQuestionAnsweringInputData,
 	DocumentQuestionAnsweringOutput,
 } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options, RequestArgs } from "../../types";
 import { base64FromBytes } from "../../utils/base64FromBytes";
@@ -19,7 +20,8 @@ export async function documentQuestionAnswering(
 	args: DocumentQuestionAnsweringArgs,
 	options?: Options
 ): Promise<DocumentQuestionAnsweringOutput[number]> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "document-question-answering");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "document-question-answering");
 	const reqArgs: RequestArgs = {
 		...args,
 		inputs: {

package/src/tasks/multimodal/visualQuestionAnswering.ts

@@ -3,6 +3,7 @@ import type {
 	VisualQuestionAnsweringInputData,
 	VisualQuestionAnsweringOutput,
 } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options, RequestArgs } from "../../types";
 import { base64FromBytes } from "../../utils/base64FromBytes";
@@ -19,7 +20,8 @@ export async function visualQuestionAnswering(
 	args: VisualQuestionAnsweringArgs,
 	options?: Options
 ): Promise<VisualQuestionAnsweringOutput[number]> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "visual-question-answering");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "visual-question-answering");
 	const reqArgs: RequestArgs = {
 		...args,
 		inputs: {

package/src/tasks/nlp/chatCompletion.ts

@@ -1,4 +1,5 @@
 import type { ChatCompletionInput, ChatCompletionOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -10,7 +11,8 @@ export async function chatCompletion(
 	args: BaseArgs & ChatCompletionInput,
 	options?: Options
 ): Promise<ChatCompletionOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "conversational");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "conversational");
 	const { data: response } = await innerRequest<ChatCompletionOutput>(args, providerHelper, {
 		...options,
 		task: "conversational",

package/src/tasks/nlp/chatCompletionStream.ts

@@ -1,4 +1,5 @@
 import type { ChatCompletionInput, ChatCompletionStreamOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerStreamingRequest } from "../../utils/request";
@@ -10,7 +11,8 @@ export async function* chatCompletionStream(
 	args: BaseArgs & ChatCompletionInput,
 	options?: Options
 ): AsyncGenerator<ChatCompletionStreamOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "conversational");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "conversational");
 	yield* innerStreamingRequest<ChatCompletionStreamOutput>(args, providerHelper, {
 		...options,
 		task: "conversational",

package/src/tasks/nlp/featureExtraction.ts

@@ -1,4 +1,5 @@
 import type { FeatureExtractionInput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -22,7 +23,8 @@ export async function featureExtraction(
 	args: FeatureExtractionArgs,
 	options?: Options
 ): Promise<FeatureExtractionOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "feature-extraction");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "feature-extraction");
 	const { data: res } = await innerRequest<FeatureExtractionOutput>(args, providerHelper, {
 		...options,
 		task: "feature-extraction",

package/src/tasks/nlp/fillMask.ts

@@ -1,4 +1,5 @@
 import type { FillMaskInput, FillMaskOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -9,7 +10,8 @@ export type FillMaskArgs = BaseArgs & FillMaskInput;
  * 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 providerHelper = getProviderHelper(args.provider ?? "hf-inference", "fill-mask");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "fill-mask");
 	const { data: res } = await innerRequest<FillMaskOutput>(args, providerHelper, {
 		...options,
 		task: "fill-mask",

package/src/tasks/nlp/questionAnswering.ts

@@ -1,4 +1,6 @@
 import type { QuestionAnsweringInput, QuestionAnsweringOutput } from "@huggingface/tasks";
+
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -12,7 +14,8 @@ export async function questionAnswering(
 	args: QuestionAnsweringArgs,
 	options?: Options
 ): Promise<QuestionAnsweringOutput[number]> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "question-answering");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "question-answering");
 	const { data: res } = await innerRequest<QuestionAnsweringOutput | QuestionAnsweringOutput[number]>(
 		args,
 		providerHelper,

package/src/tasks/nlp/sentenceSimilarity.ts

@@ -1,4 +1,5 @@
 import type { SentenceSimilarityInput, SentenceSimilarityOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -12,7 +13,8 @@ export async function sentenceSimilarity(
 	args: SentenceSimilarityArgs,
 	options?: Options
 ): Promise<SentenceSimilarityOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "sentence-similarity");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "sentence-similarity");
 	const { data: res } = await innerRequest<SentenceSimilarityOutput>(args, providerHelper, {
 		...options,
 		task: "sentence-similarity",

package/src/tasks/nlp/summarization.ts

@@ -1,4 +1,5 @@
 import type { SummarizationInput, SummarizationOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -9,7 +10,8 @@ export type SummarizationArgs = BaseArgs & SummarizationInput;
  * 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 providerHelper = getProviderHelper(args.provider ?? "hf-inference", "summarization");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "summarization");
 	const { data: res } = await innerRequest<SummarizationOutput[]>(args, providerHelper, {
 		...options,
 		task: "summarization",

package/src/tasks/nlp/tableQuestionAnswering.ts

@@ -1,4 +1,5 @@
 import type { TableQuestionAnsweringInput, TableQuestionAnsweringOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -12,7 +13,8 @@ export async function tableQuestionAnswering(
 	args: TableQuestionAnsweringArgs,
 	options?: Options
 ): Promise<TableQuestionAnsweringOutput[number]> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "table-question-answering");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "table-question-answering");
 	const { data: res } = await innerRequest<TableQuestionAnsweringOutput | TableQuestionAnsweringOutput[number]>(
 		args,
 		providerHelper,

package/src/tasks/nlp/textClassification.ts

@@ -1,4 +1,5 @@
 import type { TextClassificationInput, TextClassificationOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -12,7 +13,8 @@ export async function textClassification(
 	args: TextClassificationArgs,
 	options?: Options
 ): Promise<TextClassificationOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "text-classification");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "text-classification");
 	const { data: res } = await innerRequest<TextClassificationOutput>(args, providerHelper, {
 		...options,
 		task: "text-classification",

package/src/tasks/nlp/textGeneration.ts

@@ -1,4 +1,5 @@
 import type { TextGenerationInput, TextGenerationOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { HyperbolicTextCompletionOutput } from "../../providers/hyperbolic";
 import type { BaseArgs, Options } from "../../types";
@@ -13,7 +14,8 @@ export async function textGeneration(
 	args: BaseArgs & TextGenerationInput,
 	options?: Options
 ): Promise<TextGenerationOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "text-generation");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "text-generation");
 	const { data: response } = await innerRequest<
 		HyperbolicTextCompletionOutput | TextGenerationOutput | TextGenerationOutput[]
 	>(args, providerHelper, {

package/src/tasks/nlp/textGenerationStream.ts

@@ -1,4 +1,5 @@
 import type { TextGenerationInput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerStreamingRequest } from "../../utils/request";
@@ -90,7 +91,8 @@ export async function* textGenerationStream(
 	args: BaseArgs & TextGenerationInput,
 	options?: Options
 ): AsyncGenerator<TextGenerationStreamOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "text-generation");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "text-generation");
 	yield* innerStreamingRequest<TextGenerationStreamOutput>(args, providerHelper, {
 		...options,
 		task: "text-generation",

package/src/tasks/nlp/tokenClassification.ts

@@ -1,4 +1,5 @@
 import type { TokenClassificationInput, TokenClassificationOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -12,7 +13,8 @@ export async function tokenClassification(
 	args: TokenClassificationArgs,
 	options?: Options
 ): Promise<TokenClassificationOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "token-classification");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "token-classification");
 	const { data: res } = await innerRequest<TokenClassificationOutput[number] | TokenClassificationOutput>(
 		args,
 		providerHelper,

package/src/tasks/nlp/translation.ts

@@ -1,4 +1,5 @@
 import type { TranslationInput, TranslationOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -8,7 +9,8 @@ export type TranslationArgs = BaseArgs & TranslationInput;
  * 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 providerHelper = getProviderHelper(args.provider ?? "hf-inference", "translation");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "translation");
 	const { data: res } = await innerRequest<TranslationOutput>(args, providerHelper, {
 		...options,
 		task: "translation",

package/src/tasks/nlp/zeroShotClassification.ts

@@ -1,4 +1,5 @@
 import type { ZeroShotClassificationInput, ZeroShotClassificationOutput } from "@huggingface/tasks";
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -12,7 +13,8 @@ export async function zeroShotClassification(
 	args: ZeroShotClassificationArgs,
 	options?: Options
 ): Promise<ZeroShotClassificationOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "zero-shot-classification");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "zero-shot-classification");
 	const { data: res } = await innerRequest<ZeroShotClassificationOutput[number] | ZeroShotClassificationOutput>(
 		args,
 		providerHelper,

package/src/tasks/tabular/tabularClassification.ts

@@ -1,3 +1,4 @@
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -25,7 +26,8 @@ export async function tabularClassification(
 	args: TabularClassificationArgs,
 	options?: Options
 ): Promise<TabularClassificationOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "tabular-classification");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "tabular-classification");
 	const { data: res } = await innerRequest<TabularClassificationOutput>(args, providerHelper, {
 		...options,
 		task: "tabular-classification",

package/src/tasks/tabular/tabularRegression.ts

@@ -1,3 +1,4 @@
+import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
@@ -25,7 +26,8 @@ export async function tabularRegression(
 	args: TabularRegressionArgs,
 	options?: Options
 ): Promise<TabularRegressionOutput> {
-	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "tabular-regression");
+	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
+	const providerHelper = getProviderHelper(provider, "tabular-regression");
 	const { data: res } = await innerRequest<TabularRegressionOutput>(args, providerHelper, {
 		...options,
 		task: "tabular-regression",

package/src/types.ts

@@ -51,13 +51,18 @@ export const INFERENCE_PROVIDERS = [
 	"novita",
 	"nscale",
 	"openai",
+	"ovhcloud",
 	"replicate",
 	"sambanova",
 	"together",
 ] as const;
 
+export const PROVIDERS_OR_POLICIES = [...INFERENCE_PROVIDERS, "auto"] as const;
+
 export type InferenceProvider = (typeof INFERENCE_PROVIDERS)[number];
 
+export type InferenceProviderOrPolicy = (typeof PROVIDERS_OR_POLICIES)[number];
+
 export interface BaseArgs {
 	/**
 	 * The access token to use. Without it, you'll get rate-limited quickly.
@@ -79,18 +84,18 @@ export interface BaseArgs {
 	model?: ModelId;
 
 	/**
-	 * The URL of the endpoint to use. If not specified, will call huggingface.co/api/tasks to get the default endpoint for the task.
+	 * The URL of the endpoint to use.
 	 *
-	 * If specified, will use this URL instead of the default one.
+	 * If not specified, will call the default router.huggingface.co Inference Providers endpoint.
 	 */
 	endpointUrl?: string;
 
 	/**
 	 * Set an Inference provider to run this model on.
 	 *
-	 * Defaults to the first provider in your user settings that is compatible with this model.
+	 * Defaults to "auto" i.e. the first of the providers available for the model, sorted by the user's order in https://hf.co/settings/inference-providers.
 	 */
-	provider?: InferenceProvider;
+	provider?: InferenceProviderOrPolicy;
 }
 
 export type RequestArgs = BaseArgs &

package/src/utils/typedEntries.ts

@@ -0,0 +1,5 @@
+export function typedEntries<T extends { [s: string]: T[keyof T] } | ArrayLike<T[keyof T]>>(
+	obj: T
+): [keyof T, T[keyof T]][] {
+	return Object.entries(obj) as [keyof T, T[keyof T]][];
+}