@huggingface/inference 3.6.2 → 3.7.1

package/src/snippets/getInferenceSnippets.ts

@@ -1,15 +1,15 @@
-import type { PipelineType, WidgetType } from "@huggingface/tasks/src/pipelines.js";
-import type { ChatCompletionInputMessage, GenerationParameters } from "@huggingface/tasks/src/tasks/index.js";
+import { Template } from "@huggingface/jinja";
 import {
 	type InferenceSnippet,
 	type InferenceSnippetLanguage,
 	type ModelDataMinimal,
-	inferenceSnippetLanguages,
 	getModelInputSnippet,
+	inferenceSnippetLanguages,
 } from "@huggingface/tasks";
-import type { InferenceProvider, InferenceTask, RequestArgs } from "../types";
-import { Template } from "@huggingface/jinja";
+import type { PipelineType, WidgetType } from "@huggingface/tasks/src/pipelines.js";
+import type { ChatCompletionInputMessage, GenerationParameters } from "@huggingface/tasks/src/tasks/index.js";
 import { makeRequestOptionsFromResolvedModel } from "../lib/makeRequestOptions";
+import type { InferenceProvider, InferenceTask, RequestArgs } from "../types";
 import { templates } from "./templates.exported";
 
 const PYTHON_CLIENTS = ["huggingface_hub", "fal_client", "requests", "openai"] as const;
@@ -120,6 +120,7 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 		opts?: Record<string, unknown>
 	): InferenceSnippet[] => {
 		/// Hacky: hard-code conversational templates here
+		let task = model.pipeline_tag as InferenceTask;
 		if (
 			model.pipeline_tag &&
 			["text-generation", "image-text-to-text"].includes(model.pipeline_tag) &&
@@ -127,14 +128,20 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 		) {
 			templateName = opts?.streaming ? "conversationalStream" : "conversational";
 			inputPreparationFn = prepareConversationalInput;
+			task = "conversational";
 		}
-
 		/// Prepare inputs + make request
 		const inputs = inputPreparationFn ? inputPreparationFn(model, opts) : { inputs: getModelInputSnippet(model) };
 		const request = makeRequestOptionsFromResolvedModel(
 			providerModelId ?? model.id,
-			{ accessToken: accessToken, provider: provider, ...inputs } as RequestArgs,
-			{ chatCompletion: templateName.includes("conversational"), task: model.pipeline_tag as InferenceTask }
+			{
+				accessToken: accessToken,
+				provider: provider,
+				...inputs,
+			} as RequestArgs,
+			{
+				task: task,
+			}
 		);
 
 		/// Parse request.info.body if not a binary.
@@ -247,7 +254,7 @@ const prepareConversationalInput = (
 	return {
 		messages: opts?.messages ?? getModelInputSnippet(model),
 		...(opts?.temperature ? { temperature: opts?.temperature } : undefined),
-		max_tokens: opts?.max_tokens ?? 500,
+		max_tokens: opts?.max_tokens ?? 512,
 		...(opts?.top_p ? { top_p: opts?.top_p } : undefined),
 	};
 };

package/src/snippets/templates.exported.ts

@@ -6,7 +6,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "basicAudio": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"audio/flac\"\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});",
       "basicImage": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"image/jpeg\"\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});",
       "textToAudio": "{% if model.library_name == \"transformers\" %}\nasync function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.blob();\n    return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    // Returns a byte object of the Audio wavform. Use it directly!\n});\n{% else %}\nasync function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n    const result = await response.json();\n    return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});\n{% endif %} ",
-      "textToImage": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.blob();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    // Use image\n});",
+      "textToImage": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.blob();\n\treturn result;\n}\n\n\nquery({ {{ providerInputs.asTsString }} }).then((response) => {\n    // Use image\n});",
       "zeroShotClassification": "async function query(data) {\n    const response = await fetch(\n\t\t\"{{ fullUrl }}\",\n        {\n            headers: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n                \"Content-Type\": \"application/json\",\n            },\n            method: \"POST\",\n            body: JSON.stringify(data),\n        }\n    );\n    const result = await response.json();\n    return result;\n}\n\nquery({\n    inputs: {{ providerInputs.asObj.inputs }},\n    parameters: { candidate_labels: [\"refund\", \"legal\", \"faq\"] }\n}).then((response) => {\n    console.log(JSON.stringify(response));\n});"
     },
     "huggingface.js": {
@@ -20,7 +20,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
     },
     "openai": {
       "conversational": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ accessToken }}\",\n});\n\nconst chatCompletion = await client.chat.completions.create({\n\tmodel: \"{{ providerModelId }}\",\n{{ inputs.asTsString }}\n});\n\nconsole.log(chatCompletion.choices[0].message);",
-      "conversationalStream": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ accessToken }}\",\n});\n\nlet out = \"\";\n\nconst stream = await client.chat.completions.create({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n{{ inputs.asTsString }}\n});\n\nfor await (const chunk of stream) {\n\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}  \n}"
+      "conversationalStream": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ accessToken }}\",\n});\n\nconst stream = await client.chat.completions.create({\n    model: \"{{ providerModelId }}\",\n{{ inputs.asTsString }}\n    stream: true,\n});\n\nfor await (const chunk of stream) {\n    process.stdout.write(chunk.choices[0]?.delta?.content || \"\");\n}"
     }
   },
   "python": {

package/src/tasks/audio/audioClassification.ts

@@ -1,7 +1,7 @@
 import type { AudioClassificationInput, AudioClassificationOutput } from "@huggingface/tasks";
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
-import { request } from "../custom/request";
+import { innerRequest } from "../../utils/request";
 import type { LegacyAudioInput } from "./utils";
 import { preparePayload } from "./utils";
 
@@ -15,15 +15,12 @@ export async function audioClassification(
 	args: AudioClassificationArgs,
 	options?: Options
 ): Promise<AudioClassificationOutput> {
+	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "audio-classification");
 	const payload = preparePayload(args);
-	const res = await request<AudioClassificationOutput>(payload, {
+	const { data: res } = await innerRequest<AudioClassificationOutput>(payload, {
 		...options,
 		task: "audio-classification",
 	});
-	const isValidOutput =
-		Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
-	if (!isValidOutput) {
-		throw new InferenceOutputError("Expected Array<{label: string, score: number}>");
-	}
-	return res;
+
+	return providerHelper.getResponse(res);
 }

package/src/tasks/audio/audioToAudio.ts

@@ -1,6 +1,6 @@
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
-import { request } from "../custom/request";
+import { innerRequest } from "../../utils/request";
 import type { LegacyAudioInput } from "./utils";
 import { preparePayload } from "./utils";
 
@@ -36,34 +36,11 @@ export interface AudioToAudioOutput {
  * Example model: speechbrain/sepformer-wham does audio source separation.
  */
 export async function audioToAudio(args: AudioToAudioArgs, options?: Options): Promise<AudioToAudioOutput[]> {
+	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "audio-to-audio");
 	const payload = preparePayload(args);
-	const res = await request<AudioToAudioOutput>(payload, {
+	const { data: res } = await innerRequest<AudioToAudioOutput>(payload, {
 		...options,
 		task: "audio-to-audio",
 	});
-
-	return validateOutput(res);
-}
-
-function validateOutput(output: unknown): AudioToAudioOutput[] {
-	if (!Array.isArray(output)) {
-		throw new InferenceOutputError("Expected Array");
-	}
-	if (
-		!output.every((elem): elem is AudioToAudioOutput => {
-			return (
-				typeof elem === "object" &&
-				elem &&
-				"label" in elem &&
-				typeof elem.label === "string" &&
-				"content-type" in elem &&
-				typeof elem["content-type"] === "string" &&
-				"blob" in elem &&
-				typeof elem.blob === "string"
-			);
-		})
-	) {
-		throw new InferenceOutputError("Expected Array<{label: string, audio: Blob}>");
-	}
-	return output;
+	return providerHelper.getResponse(res);
 }

package/src/tasks/audio/automaticSpeechRecognition.ts

@@ -1,11 +1,13 @@
 import type { AutomaticSpeechRecognitionInput, AutomaticSpeechRecognitionOutput } from "@huggingface/tasks";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { FAL_AI_SUPPORTED_BLOB_TYPES } from "../../providers/fal-ai";
 import type { BaseArgs, Options, RequestArgs } from "../../types";
 import { base64FromBytes } from "../../utils/base64FromBytes";
-import { request } from "../custom/request";
+import { omit } from "../../utils/omit";
+import { innerRequest } from "../../utils/request";
 import type { LegacyAudioInput } from "./utils";
 import { preparePayload } from "./utils";
-import { omit } from "../../utils/omit";
 
 export type AutomaticSpeechRecognitionArgs = BaseArgs & (AutomaticSpeechRecognitionInput | LegacyAudioInput);
 /**
@@ -16,8 +18,9 @@ export async function automaticSpeechRecognition(
 	args: AutomaticSpeechRecognitionArgs,
 	options?: Options
 ): Promise<AutomaticSpeechRecognitionOutput> {
+	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "automatic-speech-recognition");
 	const payload = await buildPayload(args);
-	const res = await request<AutomaticSpeechRecognitionOutput>(payload, {
+	const { data: res } = await innerRequest<AutomaticSpeechRecognitionOutput>(payload, {
 		...options,
 		task: "automatic-speech-recognition",
 	});
@@ -25,11 +28,9 @@ export async function automaticSpeechRecognition(
 	if (!isValidOutput) {
 		throw new InferenceOutputError("Expected {text: string}");
 	}
-	return res;
+	return providerHelper.getResponse(res);
 }
 
-const FAL_AI_SUPPORTED_BLOB_TYPES = ["audio/mpeg", "audio/mp4", "audio/wav", "audio/x-wav"];
-
 async function buildPayload(args: AutomaticSpeechRecognitionArgs): Promise<RequestArgs> {
 	if (args.provider === "fal-ai") {
 		const blob = "data" in args && args.data instanceof Blob ? args.data : "inputs" in args ? args.inputs : undefined;

package/src/tasks/audio/textToSpeech.ts

@@ -1,8 +1,7 @@
 import type { TextToSpeechInput } from "@huggingface/tasks";
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
-import { omit } from "../../utils/omit";
-import { request } from "../custom/request";
+import { innerRequest } from "../../utils/request";
 type TextToSpeechArgs = BaseArgs & TextToSpeechInput;
 
 interface OutputUrlTextToSpeechGeneration {
@@ -13,34 +12,11 @@ interface OutputUrlTextToSpeechGeneration {
  * Recommended model: espnet/kan-bayashi_ljspeech_vits
  */
 export async function textToSpeech(args: TextToSpeechArgs, options?: Options): Promise<Blob> {
-	// Replicate models expects "text" instead of "inputs"
-	const payload =
-		args.provider === "replicate"
-			? {
-					...omit(args, ["inputs", "parameters"]),
-					...args.parameters,
-					text: args.inputs,
-			  }
-			: args;
-	const res = await request<Blob | OutputUrlTextToSpeechGeneration>(payload, {
+	const provider = args.provider ?? "hf-inference";
+	const providerHelper = getProviderHelper(provider, "text-to-speech");
+	const { data: res } = await innerRequest<Blob | OutputUrlTextToSpeechGeneration>(args, {
 		...options,
 		task: "text-to-speech",
 	});
-	if (res instanceof Blob) {
-		return res;
-	}
-	if (res && typeof res === "object") {
-		if ("output" in res) {
-			if (typeof res.output === "string") {
-				const urlResponse = await fetch(res.output);
-				const blob = await urlResponse.blob();
-				return blob;
-			} else if (Array.isArray(res.output)) {
-				const urlResponse = await fetch(res.output[0]);
-				const blob = await urlResponse.blob();
-				return blob;
-			}
-		}
-	}
-	throw new InferenceOutputError("Expected Blob or object with output");
+	return providerHelper.getResponse(res);
 }

package/src/tasks/audio/utils.ts

@@ -1,4 +1,4 @@
-import type { BaseArgs, RequestArgs } from "../../types";
+import type { BaseArgs, InferenceProvider, RequestArgs } from "../../types";
 import { omit } from "../../utils/omit";
 
 /**
@@ -6,6 +6,7 @@ import { omit } from "../../utils/omit";
  */
 export interface LegacyAudioInput {
 	data: Blob | ArrayBuffer;
+	provider?: InferenceProvider;
 }
 
 export function preparePayload(args: BaseArgs & ({ inputs: Blob } | LegacyAudioInput)): RequestArgs {

package/src/tasks/custom/request.ts

@@ -1,47 +1,20 @@
 import type { InferenceTask, Options, RequestArgs } from "../../types";
-import { makeRequestOptions } from "../../lib/makeRequestOptions";
+import { innerRequest } from "../../utils/request";
 
 /**
  * Primitive to make custom calls to the inference provider
+ * @deprecated Use specific task functions instead. This function will be removed in a future version.
  */
 export async function request<T>(
 	args: RequestArgs,
 	options?: Options & {
 		/** In most cases (unless we pass a endpointUrl) we know the task */
 		task?: InferenceTask;
-		/** Is chat completion compatible */
-		chatCompletion?: boolean;
 	}
 ): Promise<T> {
-	const { url, info } = await makeRequestOptions(args, options);
-	const response = await (options?.fetch ?? fetch)(url, info);
-
-	if (options?.retry_on_error !== false && response.status === 503) {
-		return request(args, options);
-	}
-
-	if (!response.ok) {
-		const contentType = response.headers.get("Content-Type");
-		if (["application/json", "application/problem+json"].some((ct) => contentType?.startsWith(ct))) {
-			const output = await response.json();
-			if ([400, 422, 404, 500].includes(response.status) && options?.chatCompletion) {
-				throw new Error(
-					`Server ${args.model} does not seem to support chat completion. Error: ${JSON.stringify(output.error)}`
-				);
-			}
-			if (output.error || output.detail) {
-				throw new Error(JSON.stringify(output.error ?? output.detail));
-			} else {
-				throw new Error(output);
-			}
-		}
-		const message = contentType?.startsWith("text/plain;") ? await response.text() : undefined;
-		throw new Error(message ?? "An error occurred while fetching the blob");
-	}
-
-	if (response.headers.get("Content-Type")?.startsWith("application/json")) {
-		return await response.json();
-	}
-
-	return (await response.blob()) as T;
+	console.warn(
+		"The request method is deprecated and will be removed in a future version of huggingface.js. Use specific task functions instead."
+	);
+	const result = await innerRequest<T>(args, options);
+	return result.data;
 }

package/src/tasks/custom/streamingRequest.ts

@@ -1,100 +1,18 @@
 import type { InferenceTask, Options, RequestArgs } from "../../types";
-import { makeRequestOptions } from "../../lib/makeRequestOptions";
-import type { EventSourceMessage } from "../../vendor/fetch-event-source/parse";
-import { getLines, getMessages } from "../../vendor/fetch-event-source/parse";
-
+import { innerStreamingRequest } from "../../utils/request";
 /**
  * Primitive to make custom inference calls that expect server-sent events, and returns the response through a generator
+ * @deprecated Use specific task functions instead. This function will be removed in a future version.
  */
 export async function* streamingRequest<T>(
 	args: RequestArgs,
 	options?: Options & {
 		/** In most cases (unless we pass a endpointUrl) we know the task */
 		task?: InferenceTask;
-		/** Is chat completion compatible */
-		chatCompletion?: boolean;
 	}
 ): AsyncGenerator<T> {
-	const { url, info } = await makeRequestOptions({ ...args, stream: true }, options);
-	const response = await (options?.fetch ?? fetch)(url, info);
-
-	if (options?.retry_on_error !== false && response.status === 503) {
-		return yield* streamingRequest(args, options);
-	}
-	if (!response.ok) {
-		if (response.headers.get("Content-Type")?.startsWith("application/json")) {
-			const output = await response.json();
-			if ([400, 422, 404, 500].includes(response.status) && options?.chatCompletion) {
-				throw new Error(`Server ${args.model} does not seem to support chat completion. Error: ${output.error}`);
-			}
-			if (typeof output.error === "string") {
-				throw new Error(output.error);
-			}
-			if (output.error && "message" in output.error && typeof output.error.message === "string") {
-				/// OpenAI errors
-				throw new Error(output.error.message);
-			}
-		}
-
-		throw new Error(`Server response contains error: ${response.status}`);
-	}
-	if (!response.headers.get("content-type")?.startsWith("text/event-stream")) {
-		throw new Error(
-			`Server does not support event stream content type, it returned ` + response.headers.get("content-type")
-		);
-	}
-
-	if (!response.body) {
-		return;
-	}
-
-	const reader = response.body.getReader();
-	let events: EventSourceMessage[] = [];
-
-	const onEvent = (event: EventSourceMessage) => {
-		// accumulate events in array
-		events.push(event);
-	};
-
-	const onChunk = getLines(
-		getMessages(
-			() => {},
-			() => {},
-			onEvent
-		)
+	console.warn(
+		"The streamingRequest method is deprecated and will be removed in a future version of huggingface.js. Use specific task functions instead."
 	);
-
-	try {
-		while (true) {
-			const { done, value } = await reader.read();
-			if (done) {
-				return;
-			}
-			onChunk(value);
-			for (const event of events) {
-				if (event.data.length > 0) {
-					if (event.data === "[DONE]") {
-						return;
-					}
-					const data = JSON.parse(event.data);
-					if (typeof data === "object" && data !== null && "error" in data) {
-						const errorStr =
-							typeof data.error === "string"
-								? data.error
-								: typeof data.error === "object" &&
-								    data.error &&
-								    "message" in data.error &&
-								    typeof data.error.message === "string"
-								  ? data.error.message
-								  : JSON.stringify(data.error);
-						throw new Error(`Error forwarded from backend: ` + errorStr);
-					}
-					yield data as T;
-				}
-			}
-			events = [];
-		}
-	} finally {
-		reader.releaseLock();
-	}
+	yield* innerStreamingRequest(args, options);
 }

package/src/tasks/cv/imageClassification.ts

@@ -1,7 +1,7 @@
 import type { ImageClassificationInput, ImageClassificationOutput } from "@huggingface/tasks";
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
-import { request } from "../custom/request";
+import { innerRequest } from "../../utils/request";
 import { preparePayload, type LegacyImageInput } from "./utils";
 
 export type ImageClassificationArgs = BaseArgs & (ImageClassificationInput | LegacyImageInput);
@@ -14,15 +14,11 @@ export async function imageClassification(
 	args: ImageClassificationArgs,
 	options?: Options
 ): Promise<ImageClassificationOutput> {
+	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "image-classification");
 	const payload = preparePayload(args);
-	const res = await request<ImageClassificationOutput>(payload, {
+	const { data: res } = await innerRequest<ImageClassificationOutput>(payload, {
 		...options,
 		task: "image-classification",
 	});
-	const isValidOutput =
-		Array.isArray(res) && res.every((x) => typeof x.label === "string" && typeof x.score === "number");
-	if (!isValidOutput) {
-		throw new InferenceOutputError("Expected Array<{label: string, score: number}>");
-	}
-	return res;
+	return providerHelper.getResponse(res);
 }

package/src/tasks/cv/imageSegmentation.ts

@@ -1,7 +1,7 @@
 import type { ImageSegmentationInput, ImageSegmentationOutput } from "@huggingface/tasks";
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
-import { request } from "../custom/request";
+import { innerRequest } from "../../utils/request";
 import { preparePayload, type LegacyImageInput } from "./utils";
 
 export type ImageSegmentationArgs = BaseArgs & (ImageSegmentationInput | LegacyImageInput);
@@ -14,16 +14,11 @@ export async function imageSegmentation(
 	args: ImageSegmentationArgs,
 	options?: Options
 ): Promise<ImageSegmentationOutput> {
+	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "image-segmentation");
 	const payload = preparePayload(args);
-	const res = await request<ImageSegmentationOutput>(payload, {
+	const { data: res } = await innerRequest<ImageSegmentationOutput>(payload, {
 		...options,
 		task: "image-segmentation",
 	});
-	const isValidOutput =
-		Array.isArray(res) &&
-		res.every((x) => typeof x.label === "string" && typeof x.mask === "string" && typeof x.score === "number");
-	if (!isValidOutput) {
-		throw new InferenceOutputError("Expected Array<{label: string, mask: string, score: number}>");
-	}
-	return res;
+	return providerHelper.getResponse(res);
 }

package/src/tasks/cv/imageToImage.ts

@@ -1,8 +1,8 @@
 import type { ImageToImageInput } from "@huggingface/tasks";
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options, RequestArgs } from "../../types";
 import { base64FromBytes } from "../../utils/base64FromBytes";
-import { request } from "../custom/request";
+import { innerRequest } from "../../utils/request";
 
 export type ImageToImageArgs = BaseArgs & ImageToImageInput;
 
@@ -11,6 +11,7 @@ export type ImageToImageArgs = BaseArgs & ImageToImageInput;
  * Recommended model: lllyasviel/sd-controlnet-depth
  */
 export async function imageToImage(args: ImageToImageArgs, options?: Options): Promise<Blob> {
+	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "image-to-image");
 	let reqArgs: RequestArgs;
 	if (!args.parameters) {
 		reqArgs = {
@@ -26,13 +27,9 @@ export async function imageToImage(args: ImageToImageArgs, options?: Options): P
 			),
 		};
 	}
-	const res = await request<Blob>(reqArgs, {
+	const { data: res } = await innerRequest<Blob>(reqArgs, {
 		...options,
 		task: "image-to-image",
 	});
-	const isValidOutput = res && res instanceof Blob;
-	if (!isValidOutput) {
-		throw new InferenceOutputError("Expected Blob");
-	}
-	return res;
+	return providerHelper.getResponse(res);
 }

package/src/tasks/cv/imageToText.ts

@@ -1,7 +1,7 @@
 import type { ImageToTextInput, ImageToTextOutput } from "@huggingface/tasks";
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import { getProviderHelper } from "../../lib/getProviderHelper";
 import type { BaseArgs, Options } from "../../types";
-import { request } from "../custom/request";
+import { innerRequest } from "../../utils/request";
 import type { LegacyImageInput } from "./utils";
 import { preparePayload } from "./utils";
 
@@ -10,17 +10,12 @@ export type ImageToTextArgs = BaseArgs & (ImageToTextInput | LegacyImageInput);
  * This task reads some image input and outputs the text caption.
  */
 export async function imageToText(args: ImageToTextArgs, options?: Options): Promise<ImageToTextOutput> {
+	const providerHelper = getProviderHelper(args.provider ?? "hf-inference", "image-to-text");
 	const payload = preparePayload(args);
-	const res = (
-		await request<[ImageToTextOutput]>(payload, {
-			...options,
-			task: "image-to-text",
-		})
-	)?.[0];
+	const { data: res } = await innerRequest<[ImageToTextOutput]>(payload, {
+		...options,
+		task: "image-to-text",
+	});
 
-	if (typeof res?.generated_text !== "string") {
-		throw new InferenceOutputError("Expected {generated_text: string}");
-	}
-
-	return res;
+	return providerHelper.getResponse(res[0]);
 }

package/src/tasks/cv/objectDetection.ts

@@ -1,7 +1,7 @@
-import { request } from "../custom/request";
-import type { BaseArgs, Options } from "../../types";
-import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import type { ObjectDetectionInput, ObjectDetectionOutput } from "@huggingface/tasks";
+import { getProviderHelper } from "../../lib/getProviderHelper";
+import type { BaseArgs, Options } from "../../types";
+import { innerRequest } from "../../utils/request";
 import { preparePayload, type LegacyImageInput } from "./utils";
 
 export type ObjectDetectionArgs = BaseArgs & (ObjectDetectionInput | LegacyImageInput);
@@ -11,26 +11,11 @@ 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 payload = preparePayload(args);
-	const res = await request<ObjectDetectionOutput>(payload, {
+	const { data: res } = await innerRequest<ObjectDetectionOutput>(payload, {
 		...options,
 		task: "object-detection",
 	});
-	const isValidOutput =
-		Array.isArray(res) &&
-		res.every(
-			(x) =>
-				typeof x.label === "string" &&
-				typeof x.score === "number" &&
-				typeof x.box.xmin === "number" &&
-				typeof x.box.ymin === "number" &&
-				typeof x.box.xmax === "number" &&
-				typeof x.box.ymax === "number"
-		);
-	if (!isValidOutput) {
-		throw new InferenceOutputError(
-			"Expected Array<{label:string; score:number; box:{xmin:number; ymin:number; xmax:number; ymax:number}}>"
-		);
-	}
-	return res;
+	return providerHelper.getResponse(res);
 }