@huggingface/inference 3.7.1 → 3.8.0

package/src/tasks/nlp/translation.ts

@@ -9,7 +9,7 @@ export type TranslationArgs = BaseArgs & TranslationInput;
  */
 export async function translation(args: TranslationArgs, options?: Options): Promise<TranslationOutput> {
 	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "translation");
-	const { data: res } = await innerRequest<TranslationOutput>(args, {
+	const { data: res } = await innerRequest<TranslationOutput>(args, providerHelper, {
 		...options,
 		task: "translation",
 	});

package/src/tasks/nlp/zeroShotClassification.ts

@@ -2,7 +2,6 @@ import type { ZeroShotClassificationInput, ZeroShotClassificationOutput } from "
 import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
-import { toArray } from "../../utils/toArray";
 
 export type ZeroShotClassificationArgs = BaseArgs & ZeroShotClassificationInput;
 
@@ -14,9 +13,13 @@ export async function zeroShotClassification(
 	options?: Options
 ): Promise<ZeroShotClassificationOutput> {
 	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "zero-shot-classification");
-	const { data: res } = await innerRequest<ZeroShotClassificationOutput[number] | ZeroShotClassificationOutput>(args, {
-		...options,
-		task: "zero-shot-classification",
-	});
+	const { data: res } = await innerRequest<ZeroShotClassificationOutput[number] | ZeroShotClassificationOutput>(
+		args,
+		providerHelper,
+		{
+			...options,
+			task: "zero-shot-classification",
+		}
+	);
 	return providerHelper.getResponse(res);
 }

package/src/tasks/tabular/tabularClassification.ts

@@ -26,7 +26,7 @@ export async function tabularClassification(
 	options?: Options
 ): Promise<TabularClassificationOutput> {
 	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "tabular-classification");
-	const { data: res } = await innerRequest<TabularClassificationOutput>(args, {
+	const { data: res } = await innerRequest<TabularClassificationOutput>(args, providerHelper, {
 		...options,
 		task: "tabular-classification",
 	});

package/src/tasks/tabular/tabularRegression.ts

@@ -26,7 +26,7 @@ export async function tabularRegression(
 	options?: Options
 ): Promise<TabularRegressionOutput> {
 	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "tabular-regression");
-	const { data: res } = await innerRequest<TabularRegressionOutput>(args, {
+	const { data: res } = await innerRequest<TabularRegressionOutput>(args, providerHelper, {
 		...options,
 		task: "tabular-regression",
 	});

package/src/types.ts

@@ -1,4 +1,5 @@
 import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
+import type { InferenceProviderModelMapping } from "./lib/getInferenceProviderMapping";
 
 /**
  * HF model id, like "meta-llama/Llama-3.3-70B-Instruct"
@@ -117,5 +118,6 @@ export interface UrlParams {
 export interface BodyParams<T extends Record<string, unknown> = Record<string, unknown>> {
 	args: T;
 	model: string;
+	mapping?: InferenceProviderModelMapping | undefined;
 	task?: InferenceTask;
 }

package/src/utils/request.ts

@@ -1,3 +1,4 @@
+import type { getProviderHelper } from "../lib/getProviderHelper";
 import { makeRequestOptions } from "../lib/makeRequestOptions";
 import type { InferenceTask, Options, RequestArgs } from "../types";
 import type { EventSourceMessage } from "../vendor/fetch-event-source/parse";
@@ -16,6 +17,7 @@ export interface ResponseWrapper<T> {
  */
 export async function innerRequest<T>(
 	args: RequestArgs,
+	providerHelper: ReturnType<typeof getProviderHelper>,
 	options?: Options & {
 		/** In most cases (unless we pass a endpointUrl) we know the task */
 		task?: InferenceTask;
@@ -23,13 +25,13 @@ export async function innerRequest<T>(
 		chatCompletion?: boolean;
 	}
 ): Promise<ResponseWrapper<T>> {
-	const { url, info } = await makeRequestOptions(args, options);
+	const { url, info } = await makeRequestOptions(args, providerHelper, options);
 	const response = await (options?.fetch ?? fetch)(url, info);
 
 	const requestContext: ResponseWrapper<T>["requestContext"] = { url, info };
 
 	if (options?.retry_on_error !== false && response.status === 503) {
-		return innerRequest(args, options);
+		return innerRequest(args, providerHelper, options);
 	}
 
 	if (!response.ok) {
@@ -65,6 +67,7 @@ export async function innerRequest<T>(
  */
 export async function* innerStreamingRequest<T>(
 	args: RequestArgs,
+	providerHelper: ReturnType<typeof getProviderHelper>,
 	options?: Options & {
 		/** In most cases (unless we pass a endpointUrl) we know the task */
 		task?: InferenceTask;
@@ -72,11 +75,11 @@ export async function* innerStreamingRequest<T>(
 		chatCompletion?: boolean;
 	}
 ): AsyncGenerator<T> {
-	const { url, info } = await makeRequestOptions({ ...args, stream: true }, options);
+	const { url, info } = await makeRequestOptions({ ...args, stream: true }, providerHelper, options);
 	const response = await (options?.fetch ?? fetch)(url, info);
 
 	if (options?.retry_on_error !== false && response.status === 503) {
-		return yield* innerStreamingRequest(args, options);
+		return yield* innerStreamingRequest(args, providerHelper, options);
 	}
 	if (!response.ok) {
 		if (response.headers.get("Content-Type")?.startsWith("application/json")) {

package/dist/src/lib/getProviderModelId.d.ts

@@ -1,10 +0,0 @@
-import type { InferenceProvider, InferenceTask, Options, RequestArgs } from "../types";
-export declare function getProviderModelId(params: {
-    model: string;
-    provider: InferenceProvider;
-}, args: RequestArgs, options?: {
-    task?: InferenceTask;
-    chatCompletion?: boolean;
-    fetch?: Options["fetch"];
-}): Promise<string>;
-//# sourceMappingURL=getProviderModelId.d.ts.map

package/dist/src/lib/getProviderModelId.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"getProviderModelId.d.ts","sourceRoot":"","sources":["../../../src/lib/getProviderModelId.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,iBAAiB,EAAE,aAAa,EAAW,OAAO,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;AAShG,wBAAsB,kBAAkB,CACvC,MAAM,EAAE;IACP,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,iBAAiB,CAAC;CAC5B,EACD,IAAI,EAAE,WAAW,EACjB,OAAO,GAAE;IACR,IAAI,CAAC,EAAE,aAAa,CAAC;IACrB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,KAAK,CAAC,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;CACpB,GACJ,OAAO,CAAC,MAAM,CAAC,CAoDjB"}

package/src/lib/getProviderModelId.ts

@@ -1,74 +0,0 @@
-import type { WidgetType } from "@huggingface/tasks";
-import type { InferenceProvider, InferenceTask, ModelId, Options, RequestArgs } from "../types";
-import { HF_HUB_URL } from "../config";
-import { HARDCODED_MODEL_ID_MAPPING } from "../providers/consts";
-
-type InferenceProviderMapping = Partial<
-	Record<InferenceProvider, { providerId: string; status: "live" | "staging"; task: WidgetType }>
->;
-const inferenceProviderMappingCache = new Map<ModelId, InferenceProviderMapping>();
-
-export async function getProviderModelId(
-	params: {
-		model: string;
-		provider: InferenceProvider;
-	},
-	args: RequestArgs,
-	options: {
-		task?: InferenceTask;
-		chatCompletion?: boolean;
-		fetch?: Options["fetch"];
-	} = {}
-): Promise<string> {
-	if (params.provider === "hf-inference") {
-		return params.model;
-	}
-	if (!options.task) {
-		throw new Error("task must be specified when using a third-party provider");
-	}
-	const task: WidgetType =
-		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]) {
-		return HARDCODED_MODEL_ID_MAPPING[params.provider][params.model];
-	}
-
-	let inferenceProviderMapping: InferenceProviderMapping | null;
-	if (inferenceProviderMappingCache.has(params.model)) {
-		// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
-		inferenceProviderMapping = inferenceProviderMappingCache.get(params.model)!;
-	} else {
-		inferenceProviderMapping = await (options?.fetch ?? fetch)(
-			`${HF_HUB_URL}/api/models/${params.model}?expand[]=inferenceProviderMapping`,
-			{
-				headers: args.accessToken?.startsWith("hf_") ? { Authorization: `Bearer ${args.accessToken}` } : {},
-			}
-		)
-			.then((resp) => resp.json())
-			.then((json) => json.inferenceProviderMapping)
-			.catch(() => null);
-	}
-
-	if (!inferenceProviderMapping) {
-		throw new Error(`We have not been able to find inference provider information for model ${params.model}.`);
-	}
-
-	const providerMapping = inferenceProviderMapping[params.provider];
-	if (providerMapping) {
-		if (providerMapping.task !== task) {
-			throw new Error(
-				`Model ${params.model} is not supported for task ${task} and provider ${params.provider}. Supported task: ${providerMapping.task}.`
-			);
-		}
-		if (providerMapping.status === "staging") {
-			console.warn(
-				`Model ${params.model} is in staging mode for provider ${params.provider}. Meant for test purposes only.`
-			);
-		}
-		// TODO: how is it handled server-side if model has multiple tasks (e.g. `text-generation` + `conversational`)?
-		return providerMapping.providerId;
-	}
-
-	throw new Error(`Model ${params.model} is not supported provider ${params.provider}.`);
-}