@huggingface/inference 2.6.1 → 2.6.3

package/dist/index.d.ts

@@ -24,6 +24,11 @@ export interface Options {
 	 * Custom fetch function to use instead of the default one, for example to use a proxy or edit headers.
 	 */
 	fetch?: typeof fetch;
+
+	/**
+	 * (Default: "same-origin"). String | Boolean. Credentials to use for the request. If this is a string, it will be passed straight on. If it's a boolean, true will be "include" and false will not send credentials at all.
+	 */
+	includeCredentials?: string | boolean;
 }
 
 export type InferenceTask =
@@ -174,8 +179,6 @@ export function textToSpeech(args: TextToSpeechArgs, options?: Options): Promise
 export function request<T>(
 	args: RequestArgs,
 	options?: Options & {
-		/** For internal HF use, which is why it's not exposed in {@link Options} */
-		includeCredentials?: boolean;
 		/** When a model can be used for multiple tasks, and we want to run a non-default task */
 		task?: string | InferenceTask;
 		/** To load default model if needed */
@@ -188,8 +191,6 @@ export function request<T>(
 export function streamingRequest<T>(
 	args: RequestArgs,
 	options?: Options & {
-		/** For internal HF use, which is why it's not exposed in {@link Options} */
-		includeCredentials?: boolean;
 		/** When a model can be used for multiple tasks, and we want to run a non-default task */
 		task?: string | InferenceTask;
 		/** To load default model if needed */
@@ -1054,8 +1055,6 @@ export class HfInference {
 	request<T>(
 		args: Omit<RequestArgs, 'accessToken'>,
 		options?: Options & {
-			/** For internal HF use, which is why it's not exposed in {@link Options} */
-			includeCredentials?: boolean;
 			/** When a model can be used for multiple tasks, and we want to run a non-default task */
 			task?: string | InferenceTask;
 			/** To load default model if needed */
@@ -1068,8 +1067,6 @@ export class HfInference {
 	streamingRequest<T>(
 		args: Omit<RequestArgs, 'accessToken'>,
 		options?: Options & {
-			/** For internal HF use, which is why it's not exposed in {@link Options} */
-			includeCredentials?: boolean;
 			/** When a model can be used for multiple tasks, and we want to run a non-default task */
 			task?: string | InferenceTask;
 			/** To load default model if needed */
@@ -1263,8 +1260,6 @@ export class HfInferenceEndpoint {
 	request<T>(
 		args: Omit<RequestArgs, 'accessToken' | 'model'>,
 		options?: Options & {
-			/** For internal HF use, which is why it's not exposed in {@link Options} */
-			includeCredentials?: boolean;
 			/** When a model can be used for multiple tasks, and we want to run a non-default task */
 			task?: string | InferenceTask;
 			/** To load default model if needed */
@@ -1277,8 +1272,6 @@ export class HfInferenceEndpoint {
 	streamingRequest<T>(
 		args: Omit<RequestArgs, 'accessToken' | 'model'>,
 		options?: Options & {
-			/** For internal HF use, which is why it's not exposed in {@link Options} */
-			includeCredentials?: boolean;
 			/** When a model can be used for multiple tasks, and we want to run a non-default task */
 			task?: string | InferenceTask;
 			/** To load default model if needed */

package/dist/index.js

@@ -102,7 +102,7 @@ var taskCache = /* @__PURE__ */ new Map();
 var CACHE_DURATION = 10 * 60 * 1e3;
 var MAX_CACHE_ITEMS = 1e3;
 var HF_HUB_URL = "https://huggingface.co";
-async function getDefaultTask(model, accessToken) {
+async function getDefaultTask(model, accessToken, options) {
   if (isUrl(model)) {
     return null;
   }
@@ -113,7 +113,7 @@ async function getDefaultTask(model, accessToken) {
     cachedTask = void 0;
   }
   if (cachedTask === void 0) {
-    const modelTask = await fetch(`${HF_HUB_URL}/api/models/${model}?expand[]=pipeline_tag`, {
+    const modelTask = await (options?.fetch ?? fetch)(`${HF_HUB_URL}/api/models/${model}?expand[]=pipeline_tag`, {
       headers: accessToken ? { Authorization: `Bearer ${accessToken}` } : {}
     }).then((resp) => resp.json()).then((json) => json.pipeline_tag).catch(() => null);
     if (!modelTask) {
@@ -177,6 +177,14 @@ async function makeRequestOptions(args, options) {
     }
     return `${HF_INFERENCE_API_BASE_URL}/models/${model}`;
   })();
+  let credentials;
+  if (typeof includeCredentials === "string") {
+    credentials = includeCredentials;
+  } else if (typeof includeCredentials === "boolean") {
+    credentials = includeCredentials ? "include" : void 0;
+  } else if (includeCredentials === void 0) {
+    credentials = "same-origin";
+  }
   const info = {
     headers,
     method: "POST",
@@ -184,7 +192,7 @@ async function makeRequestOptions(args, options) {
       ...otherArgs,
       options: options && otherOptions
     }),
-    credentials: includeCredentials ? "include" : "same-origin"
+    credentials
   };
   return { url, info };
 }
@@ -332,7 +340,7 @@ async function* streamingRequest(args, options) {
     }
     throw new Error(`Server response contains error: ${response.status}`);
   }
-  if (response.headers.get("content-type") !== "text/event-stream") {
+  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")
     );
@@ -590,7 +598,7 @@ async function conversational(args, options) {
 
 // src/tasks/nlp/featureExtraction.ts
 async function featureExtraction(args, options) {
-  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : void 0;
+  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : void 0;
   const res = await request(args, {
     ...options,
     taskHint: "feature-extraction",
@@ -645,7 +653,7 @@ async function questionAnswering(args, options) {
 
 // src/tasks/nlp/sentenceSimilarity.ts
 async function sentenceSimilarity(args, options) {
-  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : void 0;
+  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : void 0;
   const res = await request(args, {
     ...options,
     taskHint: "sentence-similarity",

package/dist/index.mjs

@@ -50,7 +50,7 @@ var taskCache = /* @__PURE__ */ new Map();
 var CACHE_DURATION = 10 * 60 * 1e3;
 var MAX_CACHE_ITEMS = 1e3;
 var HF_HUB_URL = "https://huggingface.co";
-async function getDefaultTask(model, accessToken) {
+async function getDefaultTask(model, accessToken, options) {
   if (isUrl(model)) {
     return null;
   }
@@ -61,7 +61,7 @@ async function getDefaultTask(model, accessToken) {
     cachedTask = void 0;
   }
   if (cachedTask === void 0) {
-    const modelTask = await fetch(`${HF_HUB_URL}/api/models/${model}?expand[]=pipeline_tag`, {
+    const modelTask = await (options?.fetch ?? fetch)(`${HF_HUB_URL}/api/models/${model}?expand[]=pipeline_tag`, {
       headers: accessToken ? { Authorization: `Bearer ${accessToken}` } : {}
     }).then((resp) => resp.json()).then((json) => json.pipeline_tag).catch(() => null);
     if (!modelTask) {
@@ -125,6 +125,14 @@ async function makeRequestOptions(args, options) {
     }
     return `${HF_INFERENCE_API_BASE_URL}/models/${model}`;
   })();
+  let credentials;
+  if (typeof includeCredentials === "string") {
+    credentials = includeCredentials;
+  } else if (typeof includeCredentials === "boolean") {
+    credentials = includeCredentials ? "include" : void 0;
+  } else if (includeCredentials === void 0) {
+    credentials = "same-origin";
+  }
   const info = {
     headers,
     method: "POST",
@@ -132,7 +140,7 @@ async function makeRequestOptions(args, options) {
       ...otherArgs,
       options: options && otherOptions
     }),
-    credentials: includeCredentials ? "include" : "same-origin"
+    credentials
   };
   return { url, info };
 }
@@ -280,7 +288,7 @@ async function* streamingRequest(args, options) {
     }
     throw new Error(`Server response contains error: ${response.status}`);
   }
-  if (response.headers.get("content-type") !== "text/event-stream") {
+  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")
     );
@@ -538,7 +546,7 @@ async function conversational(args, options) {
 
 // src/tasks/nlp/featureExtraction.ts
 async function featureExtraction(args, options) {
-  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : void 0;
+  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : void 0;
   const res = await request(args, {
     ...options,
     taskHint: "feature-extraction",
@@ -593,7 +601,7 @@ async function questionAnswering(args, options) {
 
 // src/tasks/nlp/sentenceSimilarity.ts
 async function sentenceSimilarity(args, options) {
-  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : void 0;
+  const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : void 0;
   const res = await request(args, {
     ...options,
     taskHint: "sentence-similarity",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huggingface/inference",
-  "version": "2.6.1",
+  "version": "2.6.3",
   "packageManager": "pnpm@8.3.1",
   "license": "MIT",
   "author": "Tim Mikeladze <tim.mikeladze@gmail.com>",

package/src/lib/getDefaultTask.ts

@@ -10,13 +10,21 @@ const CACHE_DURATION = 10 * 60 * 1000;
 const MAX_CACHE_ITEMS = 1000;
 export const HF_HUB_URL = "https://huggingface.co";
 
+export interface DefaultTaskOptions {
+	fetch?: typeof fetch;
+}
+
 /**
  * Get the default task. Use a LRU cache of 1000 items with 10 minutes expiration
  * to avoid making too many calls to the HF hub.
  *
  * @returns The default task for the model, or `null` if it was impossible to get it
  */
-export async function getDefaultTask(model: string, accessToken: string | undefined): Promise<string | null> {
+export async function getDefaultTask(
+	model: string,
+	accessToken: string | undefined,
+	options?: DefaultTaskOptions
+): Promise<string | null> {
 	if (isUrl(model)) {
 		return null;
 	}
@@ -30,7 +38,7 @@ export async function getDefaultTask(model: string, accessToken: string | undefi
 	}
 
 	if (cachedTask === undefined) {
-		const modelTask = await fetch(`${HF_HUB_URL}/api/models/${model}?expand[]=pipeline_tag`, {
+		const modelTask = await (options?.fetch ?? fetch)(`${HF_HUB_URL}/api/models/${model}?expand[]=pipeline_tag`, {
 			headers: accessToken ? { Authorization: `Bearer ${accessToken}` } : {},
 		})
 			.then((resp) => resp.json())

package/src/lib/makeRequestOptions.ts

@@ -18,8 +18,6 @@ export async function makeRequestOptions(
 		stream?: boolean;
 	},
 	options?: Options & {
-		/** For internal HF use, which is why it's not exposed in {@link Options} */
-		includeCredentials?: boolean;
 		/** 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 */
@@ -83,6 +81,21 @@ export async function makeRequestOptions(
 		return `${HF_INFERENCE_API_BASE_URL}/models/${model}`;
 	})();
 
+	// Let users configure credentials, or disable them all together (or keep default behavior).
+	// ---
+	// This used to be an internal property only and never exposed to users. This means that most usages will never define this value
+	// So in order to make this backwards compatible, if it's undefined we go to "same-origin" (default behaviour before).
+	// If it's a boolean and set to true then set to "include". If false, don't define credentials at all (useful for edge runtimes)
+	// Then finally, if it's a string, use it as-is.
+	let credentials: RequestCredentials | undefined;
+	if (typeof includeCredentials === "string") {
+		credentials = includeCredentials as RequestCredentials;
+	} else if (typeof includeCredentials === "boolean") {
+		credentials = includeCredentials ? "include" : undefined;
+	} else if (includeCredentials === undefined) {
+		credentials = "same-origin";
+	}
+
 	const info: RequestInit = {
 		headers,
 		method: "POST",
@@ -92,7 +105,7 @@ export async function makeRequestOptions(
 					...otherArgs,
 					options: options && otherOptions,
 			  }),
-		credentials: includeCredentials ? "include" : "same-origin",
+		credentials,
 	};
 
 	return { url, info };

package/src/tasks/custom/request.ts

@@ -7,8 +7,6 @@ import { makeRequestOptions } from "../../lib/makeRequestOptions";
 export async function request<T>(
 	args: RequestArgs,
 	options?: Options & {
-		/** For internal HF use, which is why it's not exposed in {@link Options} */
-		includeCredentials?: boolean;
 		/** When a model can be used for multiple tasks, and we want to run a non-default task */
 		task?: string | InferenceTask;
 		/** To load default model if needed */

package/src/tasks/custom/streamingRequest.ts

@@ -9,8 +9,6 @@ import { getLines, getMessages } from "../../vendor/fetch-event-source/parse";
 export async function* streamingRequest<T>(
 	args: RequestArgs,
 	options?: Options & {
-		/** For internal HF use, which is why it's not exposed in {@link Options} */
-		includeCredentials?: boolean;
 		/** When a model can be used for multiple tasks, and we want to run a non-default task */
 		task?: string | InferenceTask;
 		/** To load default model if needed */
@@ -36,7 +34,7 @@ export async function* streamingRequest<T>(
 
 		throw new Error(`Server response contains error: ${response.status}`);
 	}
-	if (response.headers.get("content-type") !== "text/event-stream") {
+	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")
 		);

package/src/tasks/nlp/featureExtraction.ts

@@ -25,7 +25,7 @@ export async function featureExtraction(
 	args: FeatureExtractionArgs,
 	options?: Options
 ): Promise<FeatureExtractionOutput> {
-	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : undefined;
+	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : undefined;
 
 	const res = await request<FeatureExtractionOutput>(args, {
 		...options,

package/src/tasks/nlp/sentenceSimilarity.ts

@@ -25,7 +25,7 @@ export async function sentenceSimilarity(
 	args: SentenceSimilarityArgs,
 	options?: Options
 ): Promise<SentenceSimilarityOutput> {
-	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken) : undefined;
+	const defaultTask = args.model ? await getDefaultTask(args.model, args.accessToken, options) : undefined;
 	const res = await request<SentenceSimilarityOutput>(args, {
 		...options,
 		taskHint: "sentence-similarity",

package/src/types.ts

@@ -24,6 +24,11 @@ export interface Options {
 	 * Custom fetch function to use instead of the default one, for example to use a proxy or edit headers.
 	 */
 	fetch?: typeof fetch;
+
+	/**
+	 * (Default: "same-origin"). String | Boolean. Credentials to use for the request. If this is a string, it will be passed straight on. If it's a boolean, true will be "include" and false will not send credentials at all.
+	 */
+	includeCredentials?: string | boolean;
 }
 
 export type InferenceTask =