Merge branch 'main' into feat/wavespeedai

Author: arabot777

README.md

@@ -97,7 +97,7 @@ You can run our packages with vanilla JS, without any bundler, by using a CDN or
 
 ```html
 <script type="module">
-    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/inference@3.13.2/+esm';
+    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/inference@3.15.0/+esm';
     import { createRepo, commit, deleteRepo, listFiles } from "https://cdn.jsdelivr.net/npm/@huggingface/[email protected]/+esm";
 </script>
 ```

packages/inference/README.md

@@ -53,7 +53,7 @@ Currently, we support the following providers:
 - [HF Inference](https://huggingface.co/docs/inference-providers/providers/hf-inference)
 - [Hyperbolic](https://hyperbolic.xyz)
 - [Nebius](https://studio.nebius.ai)
-- [Novita](https://novita.ai/?utm_source=github_huggingface&utm_medium=github_readme&utm_campaign=link)
+- [Novita](https://novita.ai)
 - [Nscale](https://nscale.com)
 - [OVHcloud](https://endpoints.ai.cloud.ovh.net/)
 - [Replicate](https://replicate.com)
@@ -97,6 +97,7 @@ Only a subset of models are supported when requesting third-party providers. You
 - [Cohere supported models](https://huggingface.co/api/partners/cohere/models)
 - [Cerebras supported models](https://huggingface.co/api/partners/cerebras/models)
 - [Groq supported models](https://console.groq.com/docs/models)
+- [Novita AI supported models](https://huggingface.co/api/partners/novita/models)
 - [Wavespeed.ai supported models](https://huggingface.co/api/partners/wavespeed-ai/models)
 
 ❗**Important note:** To be compatible, the third-party API must adhere to the "standard" shape API we expect on HF model pages for each pipeline task type.

packages/inference/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@huggingface/inference",
-	"version": "3.14.0",
+	"version": "3.15.0",
 	"packageManager": "[email protected]",
 	"license": "MIT",
 	"author": "Hugging Face and Tim Mikeladze <[email protected]>",

packages/inference/src/lib/getProviderHelper.ts

@@ -121,6 +121,7 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 	novita: {
 		conversational: new Novita.NovitaConversationalTask(),
 		"text-generation": new Novita.NovitaTextGenerationTask(),
+		"text-to-video": new Novita.NovitaTextToVideoTask(),
 	},
 	nscale: {
 		"text-to-image": new Nscale.NscaleTextToImageTask(),

packages/inference/src/package.ts

@@ -1,3 +1,3 @@
 // Generated file from package.json. Issues importing JSON directly when publishing on commonjs/ESM - see https://github.com/microsoft/TypeScript/issues/51783
-export const PACKAGE_VERSION = "3.14.0";
+export const PACKAGE_VERSION = "3.15.0";
 export const PACKAGE_NAME = "@huggingface/inference";

packages/inference/src/providers/novita.ts

@@ -16,7 +16,9 @@
  */
 import { InferenceOutputError } from "../lib/InferenceOutputError.js";
 import { isUrl } from "../lib/isUrl.js";
+import type { TextToVideoArgs } from "../tasks/index.js";
 import type { BodyParams, UrlParams } from "../types.js";
+import { delay } from "../utils/delay.js";
 import { omit } from "../utils/omit.js";
 import {
 	BaseConversationalTask,
@@ -26,11 +28,11 @@ import {
 } from "./providerHelper.js";
 
 const NOVITA_API_BASE_URL = "https://api.novita.ai";
-export interface NovitaOutput {
-	video: {
-		video_url: string;
-	};
+
+export interface NovitaAsyncAPIOutput {
+	task_id: string;
 }
+
 export class NovitaTextGenerationTask extends BaseTextGenerationTask {
 	constructor() {
 		super("novita", NOVITA_API_BASE_URL);
@@ -50,38 +52,94 @@ export class NovitaConversationalTask extends BaseConversationalTask {
 		return "/v3/openai/chat/completions";
 	}
 }
+
 export class NovitaTextToVideoTask extends TaskProviderHelper implements TextToVideoTaskHelper {
 	constructor() {
 		super("novita", NOVITA_API_BASE_URL);
 	}
 
-	makeRoute(params: UrlParams): string {
-		return `/v3/hf/${params.model}`;
+	override makeRoute(params: UrlParams): string {
+		return `/v3/async/${params.model}`;
 	}
 
-	preparePayload(params: BodyParams): Record<string, unknown> {
+	override preparePayload(params: BodyParams<TextToVideoArgs>): Record<string, unknown> {
+		const { num_inference_steps, ...restParameters } = params.args.parameters ?? {};
 		return {
 			...omit(params.args, ["inputs", "parameters"]),
-			...(params.args.parameters as Record<string, unknown>),
+			...restParameters,
+			steps: num_inference_steps,
 			prompt: params.args.inputs,
 		};
 	}
-	override async getResponse(response: NovitaOutput): Promise<Blob> {
-		const isValidOutput =
-			typeof response === "object" &&
-			!!response &&
-			"video" in response &&
-			typeof response.video === "object" &&
-			!!response.video &&
-			"video_url" in response.video &&
-			typeof response.video.video_url === "string" &&
-			isUrl(response.video.video_url);
 
-		if (!isValidOutput) {
-			throw new InferenceOutputError("Expected { video: { video_url: string } }");
+	override async getResponse(
+		response: NovitaAsyncAPIOutput,
+		url?: string,
+		headers?: Record<string, string>
+	): Promise<Blob> {
+		if (!url || !headers) {
+			throw new InferenceOutputError("URL and headers are required for text-to-video task");
 		}
+		const taskId = response.task_id;
+		if (!taskId) {
+			throw new InferenceOutputError("No task ID found in the response");
+		}
+
+		const parsedUrl = new URL(url);
+		const baseUrl = `${parsedUrl.protocol}//${parsedUrl.host}${
+			parsedUrl.host === "router.huggingface.co" ? "/novita" : ""
+		}`;
+		const resultUrl = `${baseUrl}/v3/async/task-result?task_id=${taskId}`;
+
+		let status = "";
+		let taskResult: unknown;
 
-		const urlResponse = await fetch(response.video.video_url);
-		return await urlResponse.blob();
+		while (status !== "TASK_STATUS_SUCCEED" && status !== "TASK_STATUS_FAILED") {
+			await delay(500);
+			const resultResponse = await fetch(resultUrl, { headers });
+			if (!resultResponse.ok) {
+				throw new InferenceOutputError("Failed to fetch task result");
+			}
+			try {
+				taskResult = await resultResponse.json();
+				if (
+					taskResult &&
+					typeof taskResult === "object" &&
+					"task" in taskResult &&
+					taskResult.task &&
+					typeof taskResult.task === "object" &&
+					"status" in taskResult.task &&
+					typeof taskResult.task.status === "string"
+				) {
+					status = taskResult.task.status;
+				} else {
+					throw new InferenceOutputError("Failed to get task status");
+				}
+			} catch (error) {
+				throw new InferenceOutputError("Failed to parse task result");
+			}
+		}
+
+		if (status === "TASK_STATUS_FAILED") {
+			throw new InferenceOutputError("Task failed");
+		}
+
+		if (
+			typeof taskResult === "object" &&
+			!!taskResult &&
+			"videos" in taskResult &&
+			typeof taskResult.videos === "object" &&
+			!!taskResult.videos &&
+			Array.isArray(taskResult.videos) &&
+			taskResult.videos.length > 0 &&
+			"video_url" in taskResult.videos[0] &&
+			typeof taskResult.videos[0].video_url === "string" &&
+			isUrl(taskResult.videos[0].video_url)
+		) {
+			const urlResponse = await fetch(taskResult.videos[0].video_url);
+			return await urlResponse.blob();
+		} else {
+			throw new InferenceOutputError("Expected { videos: [{ video_url: string }] }");
+		}
 	}
 }

packages/inference/src/tasks/cv/textToVideo.ts

@@ -3,7 +3,7 @@ import { resolveProvider } from "../../lib/getInferenceProviderMapping.js";
 import { getProviderHelper } from "../../lib/getProviderHelper.js";
 import { makeRequestOptions } from "../../lib/makeRequestOptions.js";
 import type { FalAiQueueOutput } from "../../providers/fal-ai.js";
-import type { NovitaOutput } from "../../providers/novita.js";
+import type { NovitaAsyncAPIOutput } from "../../providers/novita.js";
 import type { ReplicateOutput } from "../../providers/replicate.js";
 import type { BaseArgs, Options } from "../../types.js";
 import { innerRequest } from "../../utils/request.js";
@@ -15,7 +15,7 @@ export type TextToVideoOutput = Blob;
 export async function textToVideo(args: TextToVideoArgs, options?: Options): Promise<TextToVideoOutput> {
 	const provider = await resolveProvider(args.provider, args.model, args.endpointUrl);
 	const providerHelper = getProviderHelper(provider, "text-to-video");
-	const { data: response } = await innerRequest<FalAiQueueOutput | ReplicateOutput | NovitaOutput>(
+	const { data: response } = await innerRequest<FalAiQueueOutput | ReplicateOutput | NovitaAsyncAPIOutput>(
 		args,
 		providerHelper,
 		{

packages/tasks/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "@huggingface/tasks",
 	"packageManager": "[email protected]",
-	"version": "0.19.9",
+	"version": "0.19.10",
 	"description": "List of ML tasks for huggingface.co/tasks",
 	"repository": "https://github.com/huggingface/huggingface.js.git",
 	"publishConfig": {

packages/tiny-agents/README.md

@@ -85,10 +85,19 @@ npx @huggingface/tiny-agents run ./my-agent
 
 Voilà! 🔥
 
+
+## Tiny Agents collection
+
+Browse our curated collection of Tiny Agents at https://huggingface.co/datasets/tiny-agents/tiny-agents. Each agent is stored in its own subdirectory, following the structure outlined above. Running an agent from the Hub is as simple as using its `agent_id`. For example, to run the [`julien-c/flux-schnell-generator`](https://huggingface.co/datasets/tiny-agents/tiny-agents/tree/main/julien-c/flux-schnell-generator) agent:
+
+```bash
+npx @huggingface/tiny-agents run "julien-c/flux-schnell-generator"
+```
+
 > [!NOTE]
-> Note: you can open a PR in the huggingface.js repo to share your agent with the community, just upload it inside the `src/agents/` directory.
+> Want to share your own agent with the community? Submit a PR to the [Tiny Agents](https://huggingface.co/datasets/tiny-agents/tiny-agents/discussions) repository on the Hub. Your submission must include an `agent.json` file, and you can optionally add a `PROMPT.md` file. To help others understand your agent's capabilities, consider including an `EXAMPLES.md` file with sample prompts and use cases.
 
-### Advanced: Programmatic Usage
+## Advanced: Programmatic Usage
 
 ```typescript
 import { Agent } from '@huggingface/tiny-agents';

packages/tiny-agents/package.json

@@ -54,6 +54,7 @@
 		"@huggingface/inference": "workspace:^",
 		"@huggingface/mcp-client": "workspace:^",
 		"@huggingface/tasks": "workspace:^",
+		"@huggingface/hub": "workspace:^",
 		"@modelcontextprotocol/sdk": "^1.11.4",
 		"zod": "^3.25.7"
 	}