[inference] Kill the ability to override a model's task (huggingface#1199)

#### Excerpt from [internal
convo](https://huggingface.slack.com/archives/C021329G2LT/p1739370292210139):

I was wondering, do you know by any chance if, in the old
hf-inference-api, the ability to override the pipeline when calling the
API between a feature-extraction and a sentence-similarity model, is
still used?

I am thinking to kill this feature from the client libraries (it's
simpler to always assume 1 model === 1 task) and i figured you might
know if we still use this anywhere

Author: julien-c

packages/inference/src/lib/makeRequestOptions.ts

@@ -28,8 +28,6 @@ export async function makeRequestOptions(
 		stream?: boolean;
 	},
 	options?: Options & {
-		/** When a model can be used for multiple tasks, and we want to run a non-default task */
-		forceTask?: string | InferenceTask;
 		/** To load default model if needed */
 		taskHint?: InferenceTask;
 		chatCompletion?: boolean;
@@ -39,14 +37,11 @@ export async function makeRequestOptions(
 	let otherArgs = remainingArgs;
 	const provider = maybeProvider ?? "hf-inference";
 
-	const { forceTask, includeCredentials, taskHint, chatCompletion } = options ?? {};
+	const { includeCredentials, taskHint, chatCompletion } = options ?? {};
 
 	if (endpointUrl && provider !== "hf-inference") {
 		throw new Error(`Cannot use endpointUrl with a third-party provider.`);
 	}
-	if (forceTask && provider !== "hf-inference") {
-		throw new Error(`Cannot use forceTask with a third-party provider.`);
-	}
 	if (maybeModel && isUrl(maybeModel)) {
 		throw new Error(`Model URLs are no longer supported. Use endpointUrl instead.`);
 	}
@@ -77,7 +72,6 @@ export async function makeRequestOptions(
 		: makeUrl({
 				authMethod,
 				chatCompletion: chatCompletion ?? false,
-				forceTask,
 				model,
 				provider: provider ?? "hf-inference",
 				taskHint,
@@ -146,7 +140,6 @@ function makeUrl(params: {
 	model: string;
 	provider: InferenceProvider;
 	taskHint: InferenceTask | undefined;
-	forceTask?: string | InferenceTask;
 }): string {
 	if (params.authMethod === "none" && params.provider !== "hf-inference") {
 		throw new Error("Authentication is required when requesting a third-party provider. Please provide accessToken");
@@ -224,13 +217,10 @@ function makeUrl(params: {
 		}
 		default: {
 			const baseUrl = HF_HUB_INFERENCE_PROXY_TEMPLATE.replaceAll("{{PROVIDER}}", "hf-inference");
-			const url = params.forceTask
-				? `${baseUrl}/pipeline/${params.forceTask}/${params.model}`
-				: `${baseUrl}/models/${params.model}`;
 			if (params.taskHint === "text-generation" && params.chatCompletion) {
-				return url + `/v1/chat/completions`;
+				return `${baseUrl}/models/${params.model}/v1/chat/completions`;
 			}
-			return url;
+			return `${baseUrl}/models/${params.model}`;
 		}
 	}
 }

packages/inference/src/tasks/nlp/featureExtraction.ts

@@ -1,5 +1,4 @@
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
-import { getDefaultTask } from "../../lib/getDefaultTask";
 import type { BaseArgs, Options } from "../../types";
 import { request } from "../custom/request";
 
@@ -25,12 +24,9 @@ export async function featureExtraction(
 	args: FeatureExtractionArgs,
 	options?: Options
 ): Promise<FeatureExtractionOutput> {
-	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : undefined;
-
 	const res = await request<FeatureExtractionOutput>(args, {
 		...options,
 		taskHint: "feature-extraction",
-		...(defaultTask === "sentence-similarity" && { forceTask: "feature-extraction" }),
 	});
 	let isValidOutput = true;
 

packages/inference/src/tasks/nlp/sentenceSimilarity.ts

@@ -1,6 +1,5 @@
 import type { SentenceSimilarityInput, SentenceSimilarityOutput } from "@huggingface/tasks";
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
-import { getDefaultTask } from "../../lib/getDefaultTask";
 import type { BaseArgs, Options } from "../../types";
 import { request } from "../custom/request";
 import { omit } from "../../utils/omit";
@@ -14,11 +13,9 @@ export async function sentenceSimilarity(
 	args: SentenceSimilarityArgs,
 	options?: Options
 ): Promise<SentenceSimilarityOutput> {
-	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : undefined;
 	const res = await request<SentenceSimilarityOutput>(prepareInput(args), {
 		...options,
 		taskHint: "sentence-similarity",
-		...(defaultTask === "feature-extraction" && { forceTask: "sentence-similarity" }),
 	});
 
 	const isValidOutput = Array.isArray(res) && res.every((x) => typeof x === "number");

packages/inference/test/HfInference.spec.ts

@@ -351,15 +351,6 @@ describe.concurrent("HfInference", () => {
 				});
 				expect(response).toEqual(expect.arrayContaining([expect.any(Number)]));
 			});
-			it("FeatureExtraction - same model as sentence similarity", async () => {
-				const response = await hf.featureExtraction({
-					model: "sentence-transformers/paraphrase-xlm-r-multilingual-v1",
-					inputs: "That is a happy person",
-				});
-
-				expect(response.length).toBeGreaterThan(10);
-				expect(response).toEqual(expect.arrayContaining([expect.any(Number)]));
-			});
 			it("FeatureExtraction - facebook/bart-base", async () => {
 				const response = await hf.featureExtraction({
 					model: "facebook/bart-base",