@huggingface/inference 2.6.7 → 2.7.0

package/dist/src/vendor/fetch-event-source/parse.d.ts

@@ -0,0 +1,69 @@
+/**
+ This file is a part of fetch-event-source package (as of v2.0.1)
+ https://github.com/Azure/fetch-event-source/blob/v2.0.1/src/parse.ts
+
+ Full package can be used after it is made compatible with nodejs:
+ https://github.com/Azure/fetch-event-source/issues/20
+
+ Below is the fetch-event-source package license:
+
+ MIT License
+
+ Copyright (c) Microsoft Corporation.
+
+ Permission is hereby granted, free of charge, to any person obtaining a copy
+ of this software and associated documentation files (the "Software"), to deal
+ in the Software without restriction, including without limitation the rights
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ copies of the Software, and to permit persons to whom the Software is
+ furnished to do so, subject to the following conditions:
+
+ The above copyright notice and this permission notice shall be included in all
+ copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ SOFTWARE
+
+ */
+/**
+ * Represents a message sent in an event stream
+ * https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#Event_stream_format
+ */
+export interface EventSourceMessage {
+    /** The event ID to set the EventSource object's last event ID value. */
+    id: string;
+    /** A string identifying the type of event described. */
+    event: string;
+    /** The event data */
+    data: string;
+    /** The reconnection interval (in milliseconds) to wait before retrying the connection */
+    retry?: number;
+}
+/**
+ * Converts a ReadableStream into a callback pattern.
+ * @param stream The input ReadableStream.
+ * @param onChunk A function that will be called on each new byte chunk in the stream.
+ * @returns {Promise<void>} A promise that will be resolved when the stream closes.
+ */
+export declare function getBytes(stream: ReadableStream<Uint8Array>, onChunk: (arr: Uint8Array) => void): Promise<void>;
+/**
+ * Parses arbitary byte chunks into EventSource line buffers.
+ * Each line should be of the format "field: value" and ends with \r, \n, or \r\n.
+ * @param onLine A function that will be called on each new EventSource line.
+ * @returns A function that should be called for each incoming byte chunk.
+ */
+export declare function getLines(onLine: (line: Uint8Array, fieldLength: number) => void): (arr: Uint8Array) => void;
+/**
+ * Parses line buffers into EventSourceMessages.
+ * @param onId A function that will be called on each `id` field.
+ * @param onRetry A function that will be called on each `retry` field.
+ * @param onMessage A function that will be called on each message.
+ * @returns A function that should be called for each incoming line buffer.
+ */
+export declare function getMessages(onId: (id: string) => void, onRetry: (retry: number) => void, onMessage?: (msg: EventSourceMessage) => void): (line: Uint8Array, fieldLength: number) => void;
+//# sourceMappingURL=parse.d.ts.map

package/dist/src/vendor/fetch-event-source/parse.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"parse.d.ts","sourceRoot":"","sources":["../../../../src/vendor/fetch-event-source/parse.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA+BG;AAEH;;;GAGG;AACH,MAAM,WAAW,kBAAkB;IAC/B,wEAAwE;IACxE,EAAE,EAAE,MAAM,CAAC;IACX,wDAAwD;IACxD,KAAK,EAAE,MAAM,CAAC;IACd,qBAAqB;IACrB,IAAI,EAAE,MAAM,CAAC;IACb,yFAAyF;IACzF,KAAK,CAAC,EAAE,MAAM,CAAC;CAClB;AAED;;;;;GAKG;AACH,wBAAsB,QAAQ,CAAC,MAAM,EAAE,cAAc,CAAC,UAAU,CAAC,EAAE,OAAO,EAAE,CAAC,GAAG,EAAE,UAAU,KAAK,IAAI,iBAMpG;AASD;;;;;GAKG;AACH,wBAAgB,QAAQ,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,UAAU,EAAE,WAAW,EAAE,MAAM,KAAK,IAAI,SAO/C,UAAU,UA4D1C;AAED;;;;;;GAMG;AACH,wBAAgB,WAAW,CACvB,IAAI,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,IAAI,EAC1B,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,EAChC,SAAS,CAAC,EAAE,CAAC,GAAG,EAAE,kBAAkB,KAAK,IAAI,UAMhB,UAAU,eAAe,MAAM,UAmC/D"}

package/dist/src/vendor/fetch-event-source/parse.spec.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=parse.spec.d.ts.map

package/dist/src/vendor/fetch-event-source/parse.spec.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"parse.spec.d.ts","sourceRoot":"","sources":["../../../../src/vendor/fetch-event-source/parse.spec.ts"],"names":[],"mappings":""}

package/dist/test/HfInference.spec.d.ts

@@ -0,0 +1,2 @@
+import "./vcr";
+//# sourceMappingURL=HfInference.spec.d.ts.map

package/dist/test/HfInference.spec.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"HfInference.spec.d.ts","sourceRoot":"","sources":["../../test/HfInference.spec.ts"],"names":[],"mappings":"AAKA,OAAO,OAAO,CAAC"}

package/dist/test/expect-closeto.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=expect-closeto.d.ts.map

package/dist/test/expect-closeto.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"expect-closeto.d.ts","sourceRoot":"","sources":["../../test/expect-closeto.ts"],"names":[],"mappings":""}

package/dist/test/test-files.d.ts

@@ -0,0 +1,2 @@
+export declare const readTestFile: (filename: string) => Uint8Array;
+//# sourceMappingURL=test-files.d.ts.map

package/dist/test/test-files.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"test-files.d.ts","sourceRoot":"","sources":["../../test/test-files.ts"],"names":[],"mappings":"AAGA,eAAO,MAAM,YAAY,aAAc,MAAM,KAAG,UAK/C,CAAC"}

package/dist/test/vcr.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=vcr.d.ts.map

package/dist/test/vcr.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"vcr.d.ts","sourceRoot":"","sources":["../../test/vcr.ts"],"names":[],"mappings":""}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huggingface/inference",
-  "version": "2.6.7",
+  "version": "2.7.0",
   "packageManager": "pnpm@8.10.5",
   "license": "MIT",
   "author": "Tim Mikeladze <tim.mikeladze@gmail.com>",
@@ -29,23 +29,25 @@
     "src"
   ],
   "source": "src/index.ts",
-  "types": "./dist/index.d.ts",
+  "types": "./dist/src/index.d.ts",
   "main": "./dist/index.cjs",
   "module": "./dist/index.js",
   "exports": {
-    "types": "./dist/index.d.ts",
+    "types": "./dist/src/index.d.ts",
     "require": "./dist/index.cjs",
     "import": "./dist/index.js"
   },
   "type": "module",
+  "dependencies": {
+    "@huggingface/tasks": "^0.10.0"
+  },
   "devDependencies": {
-    "@types/node": "18.13.0",
-    "@huggingface/tasks": "^0.8.0"
+    "@types/node": "18.13.0"
   },
   "resolutions": {},
   "scripts": {
-    "build": "tsup src/index.ts --format cjs,esm --clean && pnpm run dts",
-    "dts": "tsx scripts/generate-dts.ts",
+    "build": "tsup src/index.ts --format cjs,esm --clean && tsc --emitDeclarationOnly --declaration",
+    "dts": "tsx scripts/generate-dts.ts && tsc --noEmit dist/index.d.ts",
     "lint": "eslint --quiet --fix --ext .cjs,.ts .",
     "lint:check": "eslint --ext .cjs,.ts .",
     "format": "prettier --write .",

package/src/HfInference.ts

@@ -14,9 +14,9 @@ type TaskWithNoAccessToken = {
 	) => ReturnType<Task[key]>;
 };
 
-type TaskWithNoAccessTokenNoModel = {
+type TaskWithNoAccessTokenNoEndpointUrl = {
 	[key in keyof Task]: (
-		args: DistributiveOmit<Parameters<Task[key]>[0], "accessToken" | "model">,
+		args: DistributiveOmit<Parameters<Task[key]>[0], "accessToken" | "endpointUrl">,
 		options?: Parameters<Task[key]>[1]
 	) => ReturnType<Task[key]>;
 };
@@ -57,7 +57,7 @@ export class HfInferenceEndpoint {
 				enumerable: false,
 				value: (params: RequestArgs, options: Options) =>
 					// eslint-disable-next-line @typescript-eslint/no-explicit-any
-					fn({ ...params, accessToken, model: endpointUrl } as any, { ...defaultOptions, ...options }),
+					fn({ ...params, accessToken, endpointUrl } as any, { ...defaultOptions, ...options }),
 			});
 		}
 	}
@@ -65,4 +65,4 @@ export class HfInferenceEndpoint {
 
 export interface HfInference extends TaskWithNoAccessToken {}
 
-export interface HfInferenceEndpoint extends TaskWithNoAccessTokenNoModel {}
+export interface HfInferenceEndpoint extends TaskWithNoAccessTokenNoEndpointUrl {}

package/src/lib/makeRequestOptions.ts

@@ -1,4 +1,5 @@
 import type { InferenceTask, Options, RequestArgs } from "../types";
+import { omit } from "../utils/omit";
 import { HF_HUB_URL } from "./getDefaultTask";
 import { isUrl } from "./isUrl";
 
@@ -22,10 +23,10 @@ export async function makeRequestOptions(
 		forceTask?: string | InferenceTask;
 		/** To load default model if needed */
 		taskHint?: InferenceTask;
+		chatCompletion?: boolean;
 	}
 ): Promise<{ url: string; info: RequestInit }> {
-	// eslint-disable-next-line @typescript-eslint/no-unused-vars
-	const { accessToken, model: _model, ...otherArgs } = args;
+	const { accessToken, endpointUrl, ...otherArgs } = args;
 	let { model } = args;
 	const {
 		forceTask: task,
@@ -34,7 +35,7 @@ export async function makeRequestOptions(
 		wait_for_model,
 		use_cache,
 		dont_load_model,
-		...otherOptions
+		chatCompletion,
 	} = options ?? {};
 
 	const headers: Record<string, string> = {};
@@ -77,11 +78,17 @@ export async function makeRequestOptions(
 		headers["X-Load-Model"] = "0";
 	}
 
-	const url = (() => {
+	let url = (() => {
+		if (endpointUrl && isUrl(model)) {
+			throw new TypeError("Both model and endpointUrl cannot be URLs");
+		}
 		if (isUrl(model)) {
+			console.warn("Using a model URL is deprecated, please use the `endpointUrl` parameter instead");
 			return model;
 		}
-
+		if (endpointUrl) {
+			return endpointUrl;
+		}
 		if (task) {
 			return `${HF_INFERENCE_API_BASE_URL}/pipeline/${task}/${model}`;
 		}
@@ -89,6 +96,10 @@ export async function makeRequestOptions(
 		return `${HF_INFERENCE_API_BASE_URL}/models/${model}`;
 	})();
 
+	if (chatCompletion && !url.endsWith("/chat/completions")) {
+		url += "/v1/chat/completions";
+	}
+
 	/**
 	 * For edge runtimes, leave 'credentials' undefined, otherwise cloudflare workers will error
 	 */
@@ -105,8 +116,7 @@ export async function makeRequestOptions(
 		body: binary
 			? args.data
 			: JSON.stringify({
-					...otherArgs,
-					options: options && otherOptions,
+					...(otherArgs.model && isUrl(otherArgs.model) ? omit(otherArgs, "model") : otherArgs),
 			  }),
 		...(credentials && { credentials }),
 		signal: options?.signal,

package/src/tasks/custom/request.ts

@@ -11,6 +11,8 @@ export async function request<T>(
 		task?: string | InferenceTask;
 		/** To load default model if needed */
 		taskHint?: InferenceTask;
+		/** Is chat completion compatible */
+		chatCompletion?: boolean;
 	}
 ): Promise<T> {
 	const { url, info } = await makeRequestOptions(args, options);
@@ -26,6 +28,9 @@ export async function request<T>(
 	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 (output.error) {
 				throw new Error(output.error);
 			}

package/src/tasks/custom/streamingRequest.ts

@@ -13,6 +13,8 @@ export async function* streamingRequest<T>(
 		task?: string | InferenceTask;
 		/** To load default model if needed */
 		taskHint?: InferenceTask;
+		/** Is chat completion compatible */
+		chatCompletion?: boolean;
 	}
 ): AsyncGenerator<T> {
 	const { url, info } = await makeRequestOptions({ ...args, stream: true }, options);
@@ -27,6 +29,9 @@ export async function* streamingRequest<T>(
 	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 (output.error) {
 				throw new Error(output.error);
 			}
@@ -67,6 +72,9 @@ export async function* streamingRequest<T>(
 			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) {
 						throw new Error(data.error);

package/src/tasks/cv/imageToImage.ts

@@ -1,7 +1,7 @@
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import type { BaseArgs, Options, RequestArgs } from "../../types";
+import { base64FromBytes } from "../../utils/base64FromBytes";
 import { request } from "../custom/request";
-import { base64FromBytes } from "../../../../shared";
 
 export type ImageToImageArgs = BaseArgs & {
 	/**

package/src/tasks/cv/zeroShotImageClassification.ts

@@ -2,7 +2,7 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import type { BaseArgs, Options } from "../../types";
 import { request } from "../custom/request";
 import type { RequestArgs } from "../../types";
-import { base64FromBytes } from "../../../../shared";
+import { base64FromBytes } from "../../utils/base64FromBytes";
 
 export type ZeroShotImageClassificationArgs = BaseArgs & {
 	inputs: {

package/src/tasks/index.ts

@@ -30,6 +30,8 @@ export * from "./nlp/textGenerationStream";
 export * from "./nlp/tokenClassification";
 export * from "./nlp/translation";
 export * from "./nlp/zeroShotClassification";
+export * from "./nlp/chatCompletion";
+export * from "./nlp/chatCompletionStream";
 
 // Multimodal tasks
 export * from "./multimodal/documentQuestionAnswering";

package/src/tasks/multimodal/documentQuestionAnswering.ts

@@ -2,8 +2,8 @@ import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import type { BaseArgs, Options } from "../../types";
 import { request } from "../custom/request";
 import type { RequestArgs } from "../../types";
-import { base64FromBytes } from "../../../../shared";
 import { toArray } from "../../utils/toArray";
+import { base64FromBytes } from "../../utils/base64FromBytes";
 
 export type DocumentQuestionAnsweringArgs = BaseArgs & {
 	inputs: {

package/src/tasks/multimodal/visualQuestionAnswering.ts

@@ -1,7 +1,7 @@
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import type { BaseArgs, Options, RequestArgs } from "../../types";
+import { base64FromBytes } from "../../utils/base64FromBytes";
 import { request } from "../custom/request";
-import { base64FromBytes } from "../../../../shared";
 
 export type VisualQuestionAnsweringArgs = BaseArgs & {
 	inputs: {

package/src/tasks/nlp/chatCompletion.ts

@@ -0,0 +1,32 @@
+import { InferenceOutputError } from "../../lib/InferenceOutputError";
+import type { BaseArgs, Options } from "../../types";
+import { request } from "../custom/request";
+import type { ChatCompletionInput, ChatCompletionOutput } from "@huggingface/tasks";
+
+/**
+ * Use the chat completion endpoint to generate a response to a prompt, using OpenAI message completion API no stream
+ */
+
+export async function chatCompletion(
+	args: BaseArgs & ChatCompletionInput,
+	options?: Options
+): Promise<ChatCompletionOutput> {
+	const res = await request<ChatCompletionOutput>(args, {
+		...options,
+		taskHint: "text-generation",
+		chatCompletion: true,
+	});
+	const isValidOutput =
+		typeof res === "object" &&
+		Array.isArray(res?.choices) &&
+		typeof res?.created === "number" &&
+		typeof res?.id === "string" &&
+		typeof res?.model === "string" &&
+		typeof res?.system_fingerprint === "string" &&
+		typeof res?.usage === "object";
+
+	if (!isValidOutput) {
+		throw new InferenceOutputError("Expected ChatCompletionOutput");
+	}
+	return res;
+}

package/src/tasks/nlp/chatCompletionStream.ts

@@ -0,0 +1,17 @@
+import type { BaseArgs, Options } from "../../types";
+import { streamingRequest } from "../custom/streamingRequest";
+import type { ChatCompletionInput, ChatCompletionStreamOutput } from "@huggingface/tasks";
+
+/**
+ * Use to continue text from a prompt. Same as `textGeneration` but returns generator that can be read one token at a time
+ */
+export async function* chatCompletionStream(
+	args: BaseArgs & ChatCompletionInput,
+	options?: Options
+): AsyncGenerator<ChatCompletionStreamOutput> {
+	yield* streamingRequest<ChatCompletionStreamOutput>(args, {
+		...options,
+		taskHint: "text-generation",
+		chatCompletion: true,
+	});
+}

package/src/tasks/nlp/textGeneration.ts

@@ -1,209 +1,9 @@
+import type { TextGenerationInput, TextGenerationOutput } from "@huggingface/tasks";
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import type { BaseArgs, Options } from "../../types";
 import { request } from "../custom/request";
 
-/**
- * Inputs for Text Generation inference
- */
-export interface TextGenerationInput {
-	/**
-	 * The text to initialize generation with
-	 */
-	inputs: string;
-	/**
-	 * Additional inference parameters
-	 */
-	parameters?: TextGenerationParameters;
-	/**
-	 * Whether to stream output tokens
-	 */
-	stream?: boolean;
-	[property: string]: unknown;
-}
-
-/**
- * Additional inference parameters
- *
- * Additional inference parameters for Text Generation
- */
-export interface TextGenerationParameters {
-	/**
-	 * The number of sampling queries to run. Only the best one (in terms of total logprob) will
-	 * be returned.
-	 */
-	best_of?: number;
-	/**
-	 * Whether or not to output decoder input details
-	 */
-	decoder_input_details?: boolean;
-	/**
-	 * Whether or not to output details
-	 */
-	details?: boolean;
-	/**
-	 * Whether to use logits sampling instead of greedy decoding when generating new tokens.
-	 */
-	do_sample?: boolean;
-	/**
-	 * The maximum number of tokens to generate.
-	 */
-	max_new_tokens?: number;
-	/**
-	 * The parameter for repetition penalty. A value of 1.0 means no penalty. See [this
-	 * paper](https://hf.co/papers/1909.05858) for more details.
-	 */
-	repetition_penalty?: number;
-	/**
-	 * Whether to prepend the prompt to the generated text.
-	 */
-	return_full_text?: boolean;
-	/**
-	 * The random sampling seed.
-	 */
-	seed?: number;
-	/**
-	 * Stop generating tokens if a member of `stop_sequences` is generated.
-	 */
-	stop_sequences?: string[];
-	/**
-	 * The value used to modulate the logits distribution.
-	 */
-	temperature?: number;
-	/**
-	 * The number of highest probability vocabulary tokens to keep for top-k-filtering.
-	 */
-	top_k?: number;
-	/**
-	 * If set to < 1, only the smallest set of most probable tokens with probabilities that add
-	 * up to `top_p` or higher are kept for generation.
-	 */
-	top_p?: number;
-	/**
-	 * Truncate input tokens to the given size.
-	 */
-	truncate?: number;
-	/**
-	 * Typical Decoding mass. See [Typical Decoding for Natural Language
-	 * Generation](https://hf.co/papers/2202.00666) for more information
-	 */
-	typical_p?: number;
-	/**
-	 * Watermarking with [A Watermark for Large Language Models](https://hf.co/papers/2301.10226)
-	 */
-	watermark?: boolean;
-	[property: string]: unknown;
-}
-
-/**
- * Outputs for Text Generation inference
- */
-export interface TextGenerationOutput {
-	/**
-	 * When enabled, details about the generation
-	 */
-	details?: TextGenerationOutputDetails;
-	/**
-	 * The generated text
-	 */
-	generated_text: string;
-	[property: string]: unknown;
-}
-
-/**
- * When enabled, details about the generation
- */
-export interface TextGenerationOutputDetails {
-	/**
-	 * Details about additional sequences when best_of is provided
-	 */
-	best_of_sequences?: TextGenerationOutputSequenceDetails[];
-	/**
-	 * The reason why the generation was stopped.
-	 */
-	finish_reason: TextGenerationFinishReason;
-	/**
-	 * The number of generated tokens
-	 */
-	generated_tokens: number;
-	prefill: TextGenerationPrefillToken[];
-	/**
-	 * The random seed used for generation
-	 */
-	seed?: number;
-	/**
-	 * The generated tokens and associated details
-	 */
-	tokens: TextGenerationOutputToken[];
-	/**
-	 * Most likely tokens
-	 */
-	top_tokens?: Array<TextGenerationOutputToken[]>;
-	[property: string]: unknown;
-}
-
-export interface TextGenerationOutputSequenceDetails {
-	finish_reason: TextGenerationFinishReason;
-	/**
-	 * The generated text
-	 */
-	generated_text: string;
-	/**
-	 * The number of generated tokens
-	 */
-	generated_tokens: number;
-	prefill: TextGenerationPrefillToken[];
-	/**
-	 * The random seed used for generation
-	 */
-	seed?: number;
-	/**
-	 * The generated tokens and associated details
-	 */
-	tokens: TextGenerationOutputToken[];
-	/**
-	 * Most likely tokens
-	 */
-	top_tokens?: Array<TextGenerationOutputToken[]>;
-	[property: string]: unknown;
-}
-
-export interface TextGenerationPrefillToken {
-	id: number;
-	logprob: number;
-	/**
-	 * The text associated with that token
-	 */
-	text: string;
-	[property: string]: unknown;
-}
-
-/**
- * Generated token.
- */
-export interface TextGenerationOutputToken {
-	id: number;
-	logprob?: number;
-	/**
-	 * Whether or not that token is a special one
-	 */
-	special: boolean;
-	/**
-	 * The text associated with that token
-	 */
-	text: string;
-	[property: string]: unknown;
-}
-
-/**
- * The reason why the generation was stopped.
- *
- * length: The generated sequence reached the maximum allowed length
- *
- * eos_token: The model generated an end-of-sentence (EOS) token
- *
- * stop_sequence: One of the sequence in stop_sequences was generated
- */
-export type TextGenerationFinishReason = "length" | "eos_token" | "stop_sequence";
+export type { TextGenerationInput, TextGenerationOutput };
 
 /**
  * Use to continue text from a prompt. This is a very generic task. Recommended model: gpt2 (it’s a simple model, but fun to play with).

package/src/tasks/nlp/textGenerationStream.ts

@@ -1,6 +1,6 @@
+import type { TextGenerationInput } from "@huggingface/tasks";
 import type { BaseArgs, Options } from "../../types";
 import { streamingRequest } from "../custom/streamingRequest";
-import type { TextGenerationInput } from "./textGeneration";
 
 export interface TextGenerationStreamToken {
 	/** Token ID from the model tokenizer */
@@ -67,6 +67,7 @@ export interface TextGenerationStreamDetails {
 }
 
 export interface TextGenerationStreamOutput {
+	index?: number;
 	/** Generated token, one at a time */
 	token: TextGenerationStreamToken;
 	/**

package/src/types.ts

@@ -1,4 +1,5 @@
 import type { PipelineType } from "@huggingface/tasks";
+import type { ChatCompletionInput } from "@huggingface/tasks";
 
 export interface Options {
 	/**
@@ -32,7 +33,7 @@ export interface Options {
 	signal?: AbortSignal;
 
 	/**
-	 * 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 (which defaults to "same-origin" inside browsers).
+	 * (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;
 }
@@ -47,15 +48,25 @@ export interface BaseArgs {
 	 */
 	accessToken?: string;
 	/**
-	 * The model to use. Can be a full URL for a dedicated inference endpoint.
+	 * The model to use.
 	 *
 	 * If not specified, will call huggingface.co/api/tasks to get the default model for the task.
+	 *
+	 * /!\ Legacy behavior allows this to be an URL, but this is deprecated and will be removed in the future.
+	 * Use the `endpointUrl` parameter instead.
 	 */
 	model?: string;
+
+	/**
+	 * The URL of the endpoint to use. If not specified, will call huggingface.co/api/tasks to get the default endpoint for the task.
+	 *
+	 * If specified, will use this URL instead of the default one.
+	 */
+	endpointUrl?: string;
 }
 
 export type RequestArgs = BaseArgs &
-	({ data: Blob | ArrayBuffer } | { inputs: unknown }) & {
+	({ data: Blob | ArrayBuffer } | { inputs: unknown } | ChatCompletionInput) & {
 		parameters?: Record<string, unknown>;
 		accessToken?: string;
 	};