Merge branch 'main' into add-provider-siliconflow

Author: jackalcooper

README.md

@@ -98,7 +98,7 @@ 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@4.11.3/+esm';
+    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/inference@4.13.0/+esm';
     import { createRepo, commit, deleteRepo, listFiles } from "https://cdn.jsdelivr.net/npm/@huggingface/hub@2.6.12/+esm";
 </script>
 ```

packages/hub/src/lib/parse-safetensors-metadata.spec.ts

@@ -143,6 +143,16 @@ describe("parseSafetensorsMetadata", () => {
 		assert.strictEqual(safetensorsShardFileInfo?.total, "00072");
 	});
 
+	it("should detect sharded safetensors filename with 6 digits", async () => {
+		const safetensorsFilename = "model-00001-of-000163.safetensors"; // https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp/blob/main/model-00001-of-000163.safetensors
+		const safetensorsShardFileInfo = parseSafetensorsShardFilename(safetensorsFilename);
+
+		assert.strictEqual(safetensorsShardFileInfo?.prefix, "model-");
+		assert.strictEqual(safetensorsShardFileInfo?.basePrefix, "model");
+		assert.strictEqual(safetensorsShardFileInfo?.shard, "00001");
+		assert.strictEqual(safetensorsShardFileInfo?.total, "000163");
+	});
+
 	it("should support sub-byte data types", async () => {
 		const newDataTypes: Array<"F4" | "F6_E2M3" | "F6_E3M2" | "E8M0"> = ["F4", "F6_E2M3", "F6_E3M2", "E8M0"];
 

packages/hub/src/lib/parse-safetensors-metadata.ts

@@ -14,7 +14,7 @@ export const SAFETENSORS_INDEX_FILE = "model.safetensors.index.json";
 export const RE_SAFETENSORS_FILE = /\.safetensors$/;
 export const RE_SAFETENSORS_INDEX_FILE = /\.safetensors\.index\.json$/;
 export const RE_SAFETENSORS_SHARD_FILE =
-	/^(?<prefix>(?<basePrefix>.*?)[_-])(?<shard>\d{5})-of-(?<total>\d{5})\.safetensors$/;
+	/^(?<prefix>(?<basePrefix>.*?)[_-])(?<shard>\d{5,6})-of-(?<total>\d{5,6})\.safetensors$/;
 export interface SafetensorsShardFileInfo {
 	prefix: string;
 	basePrefix: string;

packages/inference/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@huggingface/inference",
-	"version": "4.11.3",
+	"version": "4.13.0",
 	"packageManager": "pnpm@10.10.0",
 	"license": "MIT",
 	"author": "Hugging Face and Tim Mikeladze <tim.mikeladze@gmail.com>",

packages/inference/src/errors.ts

@@ -17,6 +17,13 @@ export class InferenceClientInputError extends InferenceClientError {
 	}
 }
 
+export class InferenceClientRoutingError extends InferenceClientError {
+	constructor(message: string) {
+		super(message);
+		this.name = "RoutingError";
+	}
+}
+
 interface HttpRequest {
 	url: string;
 	method: string;

packages/inference/src/lib/getInferenceProviderMapping.ts

@@ -124,6 +124,17 @@ export async function getInferenceProviderMapping(
 	}
 ): Promise<InferenceProviderMappingEntry | null> {
 	const logger = getLogger();
+	if (params.provider === ("auto" as InferenceProvider) && params.task === "conversational") {
+		// Special case for auto + conversational to avoid extra API calls
+		// Call directly the server-side auto router
+		return {
+			hfModelId: params.modelId,
+			provider: "auto",
+			providerId: params.modelId,
+			status: "live",
+			task: "conversational",
+		};
+	}
 	if (HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId]) {
 		return HARDCODED_MODEL_INFERENCE_MAPPING[params.provider][params.modelId];
 	}

packages/inference/src/lib/getProviderHelper.ts

@@ -182,6 +182,7 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 		"text-to-image": new Wavespeed.WavespeedAITextToImageTask(),
 		"text-to-video": new Wavespeed.WavespeedAITextToVideoTask(),
 		"image-to-image": new Wavespeed.WavespeedAIImageToImageTask(),
+		"image-to-video": new Wavespeed.WavespeedAIImageToVideoTask(),
 	},
 	"zai-org": {
 		conversational: new Zai.ZaiConversationalTask(),

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 = "4.11.3";
+export const PACKAGE_VERSION = "4.13.0";
 export const PACKAGE_NAME = "@huggingface/inference";

packages/inference/src/providers/providerHelper.ts

@@ -47,7 +47,7 @@ import type {
 	ZeroShotImageClassificationOutput,
 } from "@huggingface/tasks";
 import { HF_ROUTER_URL } from "../config.js";
-import { InferenceClientProviderOutputError } from "../errors.js";
+import { InferenceClientProviderOutputError, InferenceClientRoutingError } from "../errors.js";
 import type { AudioToAudioOutput } from "../tasks/audio/audioToAudio.js";
 import type { BaseArgs, BodyParams, HeaderParams, InferenceProvider, RequestArgs, UrlParams } from "../types.js";
 import { toArray } from "../utils/toArray.js";
@@ -62,7 +62,7 @@ import type { ImageSegmentationArgs } from "../tasks/cv/imageSegmentation.js";
 export abstract class TaskProviderHelper {
 	constructor(
 		readonly provider: InferenceProvider,
-		private baseUrl: string,
+		protected baseUrl: string,
 		readonly clientSideRoutingOnly: boolean = false
 	) {}
 
@@ -369,3 +369,16 @@ export class BaseTextGenerationTask extends TaskProviderHelper implements TextGe
 		throw new InferenceClientProviderOutputError("Expected Array<{generated_text: string}>");
 	}
 }
+
+export class AutoRouterConversationalTask extends BaseConversationalTask {
+	constructor() {
+		super("auto" as InferenceProvider, "https://router.huggingface.co");
+	}
+
+	override makeBaseUrl(params: UrlParams): string {
+		if (params.authMethod !== "hf-token") {
+			throw new InferenceClientRoutingError("Cannot select auto-router when using non-Hugging Face API key.");
+		}
+		return this.baseUrl;
+	}
+}

packages/inference/src/providers/wavespeed.ts

@@ -1,11 +1,17 @@
 import type { TextToImageArgs } from "../tasks/cv/textToImage.js";
 import type { ImageToImageArgs } from "../tasks/cv/imageToImage.js";
 import type { TextToVideoArgs } from "../tasks/cv/textToVideo.js";
+import type { ImageToVideoArgs } from "../tasks/cv/imageToVideo.js";
 import type { BodyParams, RequestArgs, UrlParams } from "../types.js";
 import { delay } from "../utils/delay.js";
 import { omit } from "../utils/omit.js";
 import { base64FromBytes } from "../utils/base64FromBytes.js";
-import type { TextToImageTaskHelper, TextToVideoTaskHelper, ImageToImageTaskHelper } from "./providerHelper.js";
+import type {
+	TextToImageTaskHelper,
+	TextToVideoTaskHelper,
+	ImageToImageTaskHelper,
+	ImageToVideoTaskHelper,
+} from "./providerHelper.js";
 import { TaskProviderHelper } from "./providerHelper.js";
 import {
 	InferenceClientInputError,
@@ -72,7 +78,9 @@ abstract class WavespeedAITask extends TaskProviderHelper {
 		return `/api/v3/${params.model}`;
 	}
 
-	preparePayload(params: BodyParams<ImageToImageArgs | TextToImageArgs | TextToVideoArgs>): Record<string, unknown> {
+	preparePayload(
+		params: BodyParams<ImageToImageArgs | TextToImageArgs | TextToVideoArgs | ImageToVideoArgs>
+	): Record<string, unknown> {
 		const payload: Record<string, unknown> = {
 			...omit(params.args, ["inputs", "parameters"]),
 			...params.args.parameters,
@@ -95,11 +103,17 @@ abstract class WavespeedAITask extends TaskProviderHelper {
 		url?: string,
 		headers?: Record<string, string>
 	): Promise<Blob> {
-		if (!headers) {
+		if (!url || !headers) {
 			throw new InferenceClientInputError("Headers are required for WaveSpeed AI API calls");
 		}
 
-		const resultUrl = response.data.urls.get;
+		const parsedUrl = new URL(url);
+		const resultPath = new URL(response.data.urls.get).pathname;
+		/// override the base url to use the router.huggingface.co if going through huggingface router
+		const baseUrl = `${parsedUrl.protocol}//${parsedUrl.host}${
+			parsedUrl.host === "router.huggingface.co" ? "/wavespeed" : ""
+		}`;
+		const resultUrl = `${baseUrl}${resultPath}`;
 
 		// Poll for results until completion
 		while (true) {
@@ -183,3 +197,19 @@ export class WavespeedAIImageToImageTask extends WavespeedAITask implements Imag
 		};
 	}
 }
+
+export class WavespeedAIImageToVideoTask extends WavespeedAITask implements ImageToVideoTaskHelper {
+	constructor() {
+		super(WAVESPEEDAI_API_BASE_URL);
+	}
+
+	async preparePayloadAsync(args: ImageToVideoArgs): Promise<RequestArgs> {
+		return {
+			...args,
+			inputs: args.parameters?.prompt,
+			image: base64FromBytes(
+				new Uint8Array(args.inputs instanceof ArrayBuffer ? args.inputs : await (args.inputs as Blob).arrayBuffer())
+			),
+		};
+	}
+}