@huggingface/inference 3.13.0 → 3.13.2

package/src/providers/hf-inference.ts

@@ -36,7 +36,7 @@ import type {
 import { HF_ROUTER_URL } from "../config";
 import { InferenceOutputError } from "../lib/InferenceOutputError";
 import type { TabularClassificationOutput } from "../tasks/tabular/tabularClassification";
-import type { BodyParams, UrlParams } from "../types";
+import type { BodyParams, RequestArgs, UrlParams } from "../types";
 import { toArray } from "../utils/toArray";
 import type {
 	AudioClassificationTaskHelper,
@@ -70,7 +70,10 @@ import type {
 } from "./providerHelper";
 
 import { TaskProviderHelper } from "./providerHelper";
-
+import { base64FromBytes } from "../utils/base64FromBytes";
+import type { ImageToImageArgs } from "../tasks/cv/imageToImage";
+import type { AutomaticSpeechRecognitionArgs } from "../tasks/audio/automaticSpeechRecognition";
+import { omit } from "../utils/omit";
 interface Base64ImageGeneration {
 	data: Array<{
 		b64_json: string;
@@ -221,6 +224,15 @@ export class HFInferenceAutomaticSpeechRecognitionTask
 	override async getResponse(response: AutomaticSpeechRecognitionOutput): Promise<AutomaticSpeechRecognitionOutput> {
 		return response;
 	}
+
+	async preparePayloadAsync(args: AutomaticSpeechRecognitionArgs): Promise<RequestArgs> {
+		return "data" in args
+			? args
+			: {
+					...omit(args, "inputs"),
+					data: args.inputs,
+			  };
+	}
 }
 
 export class HFInferenceAudioToAudioTask extends HFInferenceTask implements AudioToAudioTaskHelper {
@@ -326,6 +338,23 @@ export class HFInferenceImageToTextTask extends HFInferenceTask implements Image
 }
 
 export class HFInferenceImageToImageTask extends HFInferenceTask implements ImageToImageTaskHelper {
+	async preparePayloadAsync(args: ImageToImageArgs): Promise<RequestArgs> {
+		if (!args.parameters) {
+			return {
+				...args,
+				model: args.model,
+				data: args.inputs,
+			};
+		} else {
+			return {
+				...args,
+				inputs: base64FromBytes(
+					new Uint8Array(args.inputs instanceof ArrayBuffer ? args.inputs : await (args.inputs as Blob).arrayBuffer())
+				),
+			};
+		}
+	}
+
 	override async getResponse(response: Blob): Promise<Blob> {
 		if (response instanceof Blob) {
 			return response;

package/src/providers/providerHelper.ts

@@ -48,8 +48,10 @@ import type {
 import { HF_ROUTER_URL } from "../config";
 import { InferenceOutputError } from "../lib/InferenceOutputError";
 import type { AudioToAudioOutput } from "../tasks/audio/audioToAudio";
-import type { BaseArgs, BodyParams, HeaderParams, InferenceProvider, UrlParams } from "../types";
+import type { BaseArgs, BodyParams, HeaderParams, InferenceProvider, RequestArgs, UrlParams } from "../types";
 import { toArray } from "../utils/toArray";
+import type { ImageToImageArgs } from "../tasks/cv/imageToImage";
+import type { AutomaticSpeechRecognitionArgs } from "../tasks/audio/automaticSpeechRecognition";
 
 /**
  * Base class for task-specific provider helpers
@@ -142,6 +144,7 @@ export interface TextToVideoTaskHelper {
 export interface ImageToImageTaskHelper {
 	getResponse(response: unknown, url?: string, headers?: HeadersInit): Promise<Blob>;
 	preparePayload(params: BodyParams<ImageToImageInput & BaseArgs>): Record<string, unknown>;
+	preparePayloadAsync(args: ImageToImageArgs): Promise<RequestArgs>;
 }
 
 export interface ImageSegmentationTaskHelper {
@@ -245,6 +248,7 @@ export interface AudioToAudioTaskHelper {
 export interface AutomaticSpeechRecognitionTaskHelper {
 	getResponse(response: unknown, url?: string, headers?: HeadersInit): Promise<AutomaticSpeechRecognitionOutput>;
 	preparePayload(params: BodyParams<AutomaticSpeechRecognitionInput & BaseArgs>): Record<string, unknown> | BodyInit;
+	preparePayloadAsync(args: AutomaticSpeechRecognitionArgs): Promise<RequestArgs>;
 }
 
 export interface AudioClassificationTaskHelper {

package/src/snippets/getInferenceSnippets.ts

@@ -144,8 +144,7 @@ const snippetGenerator = (templateName: string, inputPreparationFn?: InputPrepar
 		}
 		let providerHelper: ReturnType<typeof getProviderHelper>;
 		try {
-			/// For the "auto" provider policy we use hf-inference snippets
-			providerHelper = getProviderHelper(provider === "auto" ? "hf-inference" : provider, task);
+			providerHelper = getProviderHelper(provider, task);
 		} catch (e) {
 			console.error(`Failed to get provider helper for ${provider} (${task})`, e);
 			return [];
@@ -285,6 +284,16 @@ const prepareConversationalInput = (
 	};
 };
 
+const prepareQuestionAnsweringInput = (model: ModelDataMinimal): object => {
+	const data = JSON.parse(getModelInputSnippet(model) as string);
+	return { question: data.question, context: data.context };
+};
+
+const prepareTableQuestionAnsweringInput = (model: ModelDataMinimal): object => {
+	const data = JSON.parse(getModelInputSnippet(model) as string);
+	return { query: data.query, table: JSON.stringify(data.table) };
+};
+
 const snippets: Partial<
 	Record<
 		PipelineType,
@@ -309,12 +318,12 @@ const snippets: Partial<
 	"image-to-image": snippetGenerator("imageToImage", prepareImageToImageInput),
 	"image-to-text": snippetGenerator("basicImage"),
 	"object-detection": snippetGenerator("basicImage"),
-	"question-answering": snippetGenerator("basic"),
+	"question-answering": snippetGenerator("questionAnswering", prepareQuestionAnsweringInput),
 	"sentence-similarity": snippetGenerator("basic"),
 	summarization: snippetGenerator("basic"),
 	"tabular-classification": snippetGenerator("tabular"),
 	"tabular-regression": snippetGenerator("tabular"),
-	"table-question-answering": snippetGenerator("basic"),
+	"table-question-answering": snippetGenerator("tableQuestionAnswering", prepareTableQuestionAnsweringInput),
 	"text-classification": snippetGenerator("basic"),
 	"text-generation": snippetGenerator("basic"),
 	"text-to-audio": snippetGenerator("textToAudio"),

package/src/snippets/templates.exported.ts

@@ -30,7 +30,7 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "textToImage": "{% if provider == \"fal-ai\" %}\nimport fal_client\n\n{% if providerInputs.asObj.loras is defined and providerInputs.asObj.loras != none %}\nresult = fal_client.subscribe(\n    \"{{ providerModelId }}\",\n    arguments={\n        \"prompt\": {{ inputs.asObj.inputs }},\n        \"loras\":{{ providerInputs.asObj.loras | tojson }},\n    },\n)\n{% else %}\nresult = fal_client.subscribe(\n    \"{{ providerModelId }}\",\n    arguments={\n        \"prompt\": {{ inputs.asObj.inputs }},\n    },\n)\n{% endif %} \nprint(result)\n{% endif %} "
     },
     "huggingface_hub": {
-      "basic": "result = client.{{ methodName }}(\n    inputs={{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n)",
+      "basic": "result = client.{{ methodName }}(\n    {{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n)",
       "basicAudio": "output = client.{{ methodName }}({{ inputs.asObj.inputs }}, model=\"{{ model.id }}\")",
       "basicImage": "output = client.{{ methodName }}({{ inputs.asObj.inputs }}, model=\"{{ model.id }}\")",
       "conversational": "completion = client.chat.completions.create(\n    model=\"{{ model.id }}\",\n{{ inputs.asPythonString }}\n)\n\nprint(completion.choices[0].message) ",
@@ -38,6 +38,8 @@ export const templates: Record<string, Record<string, Record<string, string>>> =
       "documentQuestionAnswering": "output = client.document_question_answering(\n    \"{{ inputs.asObj.image }}\",\n    question=\"{{ inputs.asObj.question }}\",\n    model=\"{{ model.id }}\",\n) ",
       "imageToImage": "# output is a PIL.Image object\nimage = client.image_to_image(\n    \"{{ inputs.asObj.inputs }}\",\n    prompt=\"{{ inputs.asObj.parameters.prompt }}\",\n    model=\"{{ model.id }}\",\n) ",
       "importInferenceClient": "from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n    provider=\"{{ provider }}\",\n    api_key=\"{{ accessToken }}\",\n{% if billTo %}\n    bill_to=\"{{ billTo }}\",\n{% endif %}\n)",
+      "questionAnswering": "answer = client.question_answering(\n    question=\"{{ inputs.asObj.question }}\",\n    context=\"{{ inputs.asObj.context }}\",\n    model=\"{{ model.id }}\",\n) ",
+      "tableQuestionAnswering": "answer = client.question_answering(\n    query=\"{{ inputs.asObj.query }}\",\n    table={{ inputs.asObj.table }},\n    model=\"{{ model.id }}\",\n) ",
       "textToImage": "# output is a PIL.Image object\nimage = client.text_to_image(\n    {{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n) ",
       "textToSpeech": "# audio is returned as bytes\naudio = client.text_to_speech(\n    {{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n) \n",
       "textToVideo": "video = client.text_to_video(\n    {{ inputs.asObj.inputs }},\n    model=\"{{ model.id }}\",\n) "

package/src/tasks/audio/automaticSpeechRecognition.ts

@@ -2,13 +2,9 @@ import type { AutomaticSpeechRecognitionInput, AutomaticSpeechRecognitionOutput
 import { resolveProvider } from "../../lib/getInferenceProviderMapping";
 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 { omit } from "../../utils/omit";
+import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
 import type { LegacyAudioInput } from "./utils";
-import { preparePayload } from "./utils";
 
 export type AutomaticSpeechRecognitionArgs = BaseArgs & (AutomaticSpeechRecognitionInput | LegacyAudioInput);
 /**
@@ -21,7 +17,7 @@ export async function automaticSpeechRecognition(
 ): Promise<AutomaticSpeechRecognitionOutput> {
 	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
 	const providerHelper = getProviderHelper(provider, "automatic-speech-recognition");
-	const payload = await buildPayload(args);
+	const payload = await providerHelper.preparePayloadAsync(args);
 	const { data: res } = await innerRequest<AutomaticSpeechRecognitionOutput>(payload, providerHelper, {
 		...options,
 		task: "automatic-speech-recognition",
@@ -32,29 +28,3 @@ export async function automaticSpeechRecognition(
 	}
 	return providerHelper.getResponse(res);
 }
-
-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;
-		const contentType = blob?.type;
-		if (!contentType) {
-			throw new Error(
-				`Unable to determine the input's content-type. Make sure your are passing a Blob when using provider fal-ai.`
-			);
-		}
-		if (!FAL_AI_SUPPORTED_BLOB_TYPES.includes(contentType)) {
-			throw new Error(
-				`Provider fal-ai does not support blob type ${contentType} - supported content types are: ${FAL_AI_SUPPORTED_BLOB_TYPES.join(
-					", "
-				)}`
-			);
-		}
-		const base64audio = base64FromBytes(new Uint8Array(await blob.arrayBuffer()));
-		return {
-			...("data" in args ? omit(args, "data") : omit(args, "inputs")),
-			audio_url: `data:${contentType};base64,${base64audio}`,
-		};
-	} else {
-		return preparePayload(args);
-	}
-}

package/src/tasks/cv/imageToImage.ts

@@ -1,8 +1,7 @@
 import type { ImageToImageInput } 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";
+import type { BaseArgs, Options } from "../../types";
 import { innerRequest } from "../../utils/request";
 
 export type ImageToImageArgs = BaseArgs & ImageToImageInput;
@@ -14,22 +13,8 @@ export type ImageToImageArgs = BaseArgs & ImageToImageInput;
 export async function imageToImage(args: ImageToImageArgs, options?: Options): Promise<Blob> {
 	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
 	const providerHelper = getProviderHelper(provider, "image-to-image");
-	let reqArgs: RequestArgs;
-	if (!args.parameters) {
-		reqArgs = {
-			accessToken: args.accessToken,
-			model: args.model,
-			data: args.inputs,
-		};
-	} else {
-		reqArgs = {
-			...args,
-			inputs: base64FromBytes(
-				new Uint8Array(args.inputs instanceof ArrayBuffer ? args.inputs : await args.inputs.arrayBuffer())
-			),
-		};
-	}
-	const { data: res } = await innerRequest<Blob>(reqArgs, providerHelper, {
+	const payload = await providerHelper.preparePayloadAsync(args);
+	const { data: res } = await innerRequest<Blob>(payload, providerHelper, {
 		...options,
 		task: "image-to-image",
 	});