Merge remote-tracking branch 'upstream/main' into feat/papersearch

Author: evalstate

packages/tasks/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "@huggingface/tasks",
 	"packageManager": "[email protected]",
-	"version": "0.19.6",
+	"version": "0.19.7",
 	"description": "List of ML tasks for huggingface.co/tasks",
 	"repository": "https://github.com/huggingface/huggingface.js.git",
 	"publishConfig": {

packages/tasks/src/model-libraries-snippets.ts

@@ -593,6 +593,39 @@ export const keras_hub = (model: ModelData): string[] => {
 	return snippets;
 };
 
+export const kimi_audio = (model: ModelData): string[] => [
+	`# Example usage for KimiAudio
+# pip install git+https://github.com/MoonshotAI/Kimi-Audio.git
+
+from kimia_infer.api.kimia import KimiAudio
+
+model = KimiAudio(model_path="${model.id}", load_detokenizer=True)
+
+sampling_params = {
+    "audio_temperature": 0.8,
+    "audio_top_k": 10,
+    "text_temperature": 0.0,
+    "text_top_k": 5,
+}
+
+# For ASR
+asr_audio = "asr_example.wav"
+messages_asr = [
+    {"role": "user", "message_type": "text", "content": "Please transcribe the following audio:"},
+    {"role": "user", "message_type": "audio", "content": asr_audio}
+]
+_, text = model.generate(messages_asr, **sampling_params, output_type="text")
+print(text)
+
+# For Q&A
+qa_audio = "qa_example.wav"
+messages_conv = [{"role": "user", "message_type": "audio", "content": qa_audio}]
+wav, text = model.generate(messages_conv, **sampling_params, output_type="both")
+sf.write("output_audio.wav", wav.cpu().view(-1).numpy(), 24000)
+print(text)
+`,
+];
+
 export const lightning_ir = (model: ModelData): string[] => {
 	if (model.tags.includes("bi-encoder")) {
 		return [

packages/tasks/src/model-libraries.ts

@@ -473,6 +473,13 @@ export const MODEL_LIBRARIES_UI_ELEMENTS = {
 		snippets: snippets.keras_hub,
 		filter: true,
 	},
+	"kimi-audio": {
+		prettyLabel: "KimiAudio",
+		repoName: "KimiAudio",
+		repoUrl: "https://github.com/MoonshotAI/Kimi-Audio",
+		snippets: snippets.kimi_audio,
+		filter: false,
+	},
 	k2: {
 		prettyLabel: "K2",
 		repoName: "k2",

packages/tiny-agents/README.md

@@ -16,7 +16,6 @@ pnpm add @huggingface/tiny-agents
 
 ```bash
 npx @huggingface/tiny-agents [command] "agent/id"
-
 ```
 
 ```
@@ -41,8 +40,8 @@ touch my-agent/agent.json
 
 ```json
 {
-	"model": "Qwen/Qwen2.5-72B-Instruct", // model id
-	"provider": "nebius", // or you can also use a local endpoint base url with `endpointUrl`
+	"model": "Qwen/Qwen2.5-72B-Instruct",
+	"provider": "nebius",
 	"servers": [
 		{
 			"type": "stdio",
@@ -55,6 +54,25 @@ touch my-agent/agent.json
 }
 ```
 
+Or using a local or remote endpoint URL:
+
+```json
+{
+	"model": "Qwen/Qwen3-32B",
+	"endpointUrl": "http://localhost:1234/v1",
+	"servers": [
+		{
+			"type": "stdio",
+			"config": {
+				"command": "npx",
+				"args": ["@playwright/mcp@latest"]
+			}
+		}
+	]
+}
+
+```
+
 Where `servers` is a list of MCP servers (we support Stdio, SSE, and HTTP servers).
 
 Optionally, you can add a `PROMPT.md` file to override the default Agent prompt.

packages/tiny-agents/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "@huggingface/tiny-agents",
 	"packageManager": "[email protected]",
-	"version": "0.2.2",
+	"version": "0.2.3",
 	"description": "Lightweight, composable agents for AI applications",
 	"repository": "https://github.com/huggingface/huggingface.js.git",
 	"publishConfig": {

packages/tiny-agents/src/cli.ts

@@ -114,11 +114,17 @@ async function main() {
 
 	const { configJson, prompt } = await loadConfigFrom(loadFrom);
 
-	const ConfigSchema = z.object({
-		model: z.string(),
-		provider: z.enum(PROVIDERS_OR_POLICIES),
-		servers: z.array(ServerConfigSchema),
-	});
+	const ConfigSchema = z
+		.object({
+			model: z.string(),
+			provider: z.enum(PROVIDERS_OR_POLICIES).optional(),
+			endpointUrl: z.string().optional(),
+			apiKey: z.string().optional(),
+			servers: z.array(ServerConfigSchema),
+		})
+		.refine((data) => data.provider !== undefined || data.endpointUrl !== undefined, {
+			message: "At least one of 'provider' or 'endpointUrl' is required",
+		});
 
 	let config: z.infer<typeof ConfigSchema>;
 	try {
@@ -129,13 +135,24 @@ async function main() {
 		process.exit(1);
 	}
 
-	const agent = new Agent({
-		provider: config.provider,
-		model: config.model,
-		apiKey: process.env.HF_TOKEN ?? "",
-		servers: config.servers,
-		prompt,
-	});
+	const agent = new Agent(
+		config.endpointUrl
+			? {
+					endpointUrl: config.endpointUrl,
+					model: config.model,
+					apiKey: config.apiKey ?? process.env.API_KEY ?? process.env.HF_TOKEN,
+					servers: config.servers,
+					prompt,
+			  }
+			: {
+					// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+					provider: config.provider!,
+					model: config.model,
+					apiKey: config.apiKey ?? process.env.API_KEY ?? process.env.HF_TOKEN,
+					servers: config.servers,
+					prompt,
+			  }
+	);
 
 	if (command === "serve") {
 		error(`Serve is not implemented yet, coming soon!`);