Merge branch 'main' into add-contexttab

Wauplin · web-flow · commit 4a6a7f16598e · 2025-06-18T17:47:28.000+02:00
diff --git a/README.md b/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@4.0.4/+esm';
+    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/inference@4.0.6/+esm';
     import { createRepo, commit, deleteRepo, listFiles } from "https://cdn.jsdelivr.net/npm/@huggingface/hub@2.2.0/+esm";
 </script>
 ```
diff --git a/packages/inference/package.json b/packages/inference/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@huggingface/inference",
-	"version": "4.0.4",
+	"version": "4.0.6",
 	"packageManager": "pnpm@10.10.0",
 	"license": "MIT",
 	"author": "Hugging Face and Tim Mikeladze <tim.mikeladze@gmail.com>",
diff --git a/packages/inference/src/lib/getInferenceProviderMapping.ts b/packages/inference/src/lib/getInferenceProviderMapping.ts
@@ -2,23 +2,12 @@ import type { WidgetType } from "@huggingface/tasks";
 import { HF_HUB_URL } from "../config.js";
 import { HARDCODED_MODEL_INFERENCE_MAPPING } from "../providers/consts.js";
 import { EQUIVALENT_SENTENCE_TRANSFORMERS_TASKS } from "../providers/hf-inference.js";
-import type { InferenceProvider, InferenceProviderOrPolicy, ModelId } from "../types.js";
+import type { InferenceProvider, InferenceProviderMappingEntry, InferenceProviderOrPolicy, ModelId } from "../types.js";
 import { typedInclude } from "../utils/typedInclude.js";
 import { InferenceClientHubApiError, InferenceClientInputError } from "../errors.js";
 
 export const inferenceProviderMappingCache = new Map<ModelId, InferenceProviderMappingEntry[]>();
 
-export interface InferenceProviderMappingEntry {
-	adapter?: string;
-	adapterWeightsPath?: string;
-	hfModelId: ModelId;
-	provider: string;
-	providerId: string;
-	status: "live" | "staging";
-	task: WidgetType;
-	type?: "single-model" | "tag-filter";
-}
-
 /**
  * Normalize inferenceProviderMapping to always return an array format.
  * This provides backward and forward compatibility for the API changes.
diff --git a/packages/inference/src/lib/makeRequestOptions.ts b/packages/inference/src/lib/makeRequestOptions.ts
@@ -1,7 +1,6 @@
 import { HF_HEADER_X_BILL_TO, HF_HUB_URL } from "../config.js";
 import { PACKAGE_NAME, PACKAGE_VERSION } from "../package.js";
-import type { InferenceTask, Options, RequestArgs } from "../types.js";
-import type { InferenceProviderMappingEntry } from "./getInferenceProviderMapping.js";
+import type { InferenceTask, InferenceProviderMappingEntry, Options, RequestArgs } from "../types.js";
 import { getInferenceProviderMapping } from "./getInferenceProviderMapping.js";
 import type { getProviderHelper } from "./getProviderHelper.js";
 import { isUrl } from "./isUrl.js";
diff --git a/packages/inference/src/package.ts b/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 = "4.0.4";
+export const PACKAGE_VERSION = "4.0.6";
 export const PACKAGE_NAME = "@huggingface/inference";
diff --git a/packages/inference/src/providers/consts.ts b/packages/inference/src/providers/consts.ts
@@ -1,5 +1,4 @@
-import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
-import type { InferenceProvider } from "../types.js";
+import type { InferenceProvider, InferenceProviderMappingEntry } from "../types.js";
 import { type ModelId } from "../types.js";
 
 /**
diff --git a/packages/inference/src/snippets/getInferenceSnippets.ts b/packages/inference/src/snippets/getInferenceSnippets.ts
@@ -8,10 +8,9 @@ import {
 } from "@huggingface/tasks";
 import type { PipelineType, WidgetType } from "@huggingface/tasks";
 import type { ChatCompletionInputMessage, GenerationParameters } from "@huggingface/tasks";
-import type { InferenceProviderMappingEntry } from "../lib/getInferenceProviderMapping.js";
 import { getProviderHelper } from "../lib/getProviderHelper.js";
 import { makeRequestOptionsFromResolvedModel } from "../lib/makeRequestOptions.js";
-import type { InferenceProviderOrPolicy, InferenceTask, RequestArgs } from "../types.js";
+import type { InferenceProviderMappingEntry, InferenceProviderOrPolicy, InferenceTask, RequestArgs } from "../types.js";
 import { templates } from "./templates.exported.js";
 
 export type InferenceSnippetOptions = {
@@ -466,7 +465,7 @@ function replaceAccessTokenPlaceholder(
 		!endpointUrl && // custom endpointUrl => use a generic API_TOKEN
 		(provider == "hf-inference" || // hf-inference provider => use $HF_TOKEN
 			(!directRequest && // if explicit directRequest => use provider-specific token
-				(!snippet.includes("https://") || // no URL provided => using a client => use $HF_TOKEN
+				(snippet.includes("InferenceClient") || // using a client => use $HF_TOKEN
 					snippet.includes("https://router.huggingface.co")))); // explicit routed request => use $HF_TOKEN
 	const accessTokenEnvVar = useHfToken
 		? "HF_TOKEN" // e.g. routed request or hf-inference
diff --git a/packages/inference/src/types.ts b/packages/inference/src/types.ts
@@ -1,5 +1,4 @@
-import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
-import type { InferenceProviderMappingEntry } from "./lib/getInferenceProviderMapping.js";
+import type { ChatCompletionInput, PipelineType, WidgetType } from "@huggingface/tasks";
 
 /**
  * HF model id, like "meta-llama/Llama-3.3-70B-Instruct"
@@ -63,6 +62,17 @@ export type InferenceProvider = (typeof INFERENCE_PROVIDERS)[number];
 
 export type InferenceProviderOrPolicy = (typeof PROVIDERS_OR_POLICIES)[number];
 
+export interface InferenceProviderMappingEntry {
+	adapter?: string;
+	adapterWeightsPath?: string;
+	hfModelId: ModelId;
+	provider: string;
+	providerId: string;
+	status: "live" | "staging";
+	task: WidgetType;
+	type?: "single-model" | "tag-filter";
+}
+
 export interface BaseArgs {
 	/**
 	 * The access token to use. Without it, you'll get rate-limited quickly.
diff --git a/packages/jinja/test/templates.test.js b/packages/jinja/test/templates.test.js
@@ -4337,7 +4337,7 @@ describe("Templates", () => {
 		// TODO add failure cases
 	});
 
-	describe("Parsing and intepretation", () => {
+	describe("Parsing and interpretation", () => {
 		describe("should interpret an AST", () => {
 			for (const [name, text] of Object.entries(TEST_PARSED)) {
 				const ast = parse(text);
diff --git a/packages/languages/README.md b/packages/languages/README.md
@@ -21,6 +21,6 @@ import { language, wikiLink, type Language } from "https://esm.sh/@huggingface/l
 import { language, wikiLink, type Language } from "npm:@huggingface/languages"
 ```
 
-Check out the [full documentation](https://huggingface.co/docs/huggingface.js/languages/README).
+Check out the [full documentation](https://huggingface.co/docs/huggingface.js/index).
 
 Acknowledging Loïck Bourdois (https://github.com/lbourdois)'s help with this package and more generally supporting languages on the HF Hub.
diff --git a/packages/mcp-client/src/McpClient.ts b/packages/mcp-client/src/McpClient.ts
@@ -186,8 +186,6 @@ export class McpClient {
 		for (const toolCall of Object.values(finalToolCalls)) {
 			const toolName = toolCall.function.name ?? "unknown";
 			/// TODO(Fix upstream type so this is always a string)^
-			const toolArgs = toolCall.function.arguments === "" ? {} : JSON.parse(toolCall.function.arguments);
-
 			const toolMessage: ChatCompletionInputMessageTool = {
 				role: "tool",
 				tool_call_id: toolCall.id,
@@ -198,11 +196,29 @@ export class McpClient {
 				messages.push(toolMessage);
 				return yield toolMessage;
 			}
+			let toolArgs: Record<string, unknown> = {};
+			try {
+				toolArgs = toolCall.function.arguments === "" ? {} : JSON.parse(toolCall.function.arguments);
+			} catch (error) {
+				if (error instanceof SyntaxError) {
+					toolMessage.content = `Invalid JSON generated by the model: ${error.message}`;
+					messages.push(toolMessage);
+					yield toolMessage;
+					continue;
+				} else {
+					throw error;
+				}
+			}
+
 			/// Get the appropriate session for this tool
 			const client = this.clients.get(toolName);
 			if (client) {
-				const result = await client.callTool({ name: toolName, arguments: toolArgs, signal: opts.abortSignal });
-				toolMessage.content = ResultFormatter.format(result);
+				try {
+					const result = await client.callTool({ name: toolName, arguments: toolArgs, signal: opts.abortSignal });
+					toolMessage.content = ResultFormatter.format(result);
+				} catch (error) {
+					toolMessage.content = `Error: MCP tool call failed with error message: ${error}`;
+				}
 			} else {
 				toolMessage.content = `Error: No session found for tool: ${toolName}`;
 			}
diff --git a/packages/tasks-gen/snippets-fixtures/conversational-vlm-non-stream/js/huggingface.js/0.fireworks-ai.js b/packages/tasks-gen/snippets-fixtures/conversational-vlm-non-stream/js/huggingface.js/0.fireworks-ai.js
@@ -1,6 +1,6 @@
 import { InferenceClient } from "@huggingface/inference";
 
-const client = new InferenceClient(process.env.FIREWORKS_AI_API_KEY);
+const client = new InferenceClient(process.env.HF_TOKEN);
 
 const chatCompletion = await client.chatCompletion({
     provider: "fireworks-ai",
diff --git a/packages/tasks-gen/snippets-fixtures/conversational-vlm-non-stream/python/huggingface_hub/0.fireworks-ai.py b/packages/tasks-gen/snippets-fixtures/conversational-vlm-non-stream/python/huggingface_hub/0.fireworks-ai.py
@@ -3,7 +3,7 @@
 
 client = InferenceClient(
     provider="fireworks-ai",
-    api_key=os.environ["FIREWORKS_AI_API_KEY"],
+    api_key=os.environ["HF_TOKEN"],
 )
 
 completion = client.chat.completions.create(
diff --git a/packages/tasks-gen/snippets-fixtures/conversational-vlm-stream/js/huggingface.js/0.fireworks-ai.js b/packages/tasks-gen/snippets-fixtures/conversational-vlm-stream/js/huggingface.js/0.fireworks-ai.js
@@ -1,6 +1,6 @@
 import { InferenceClient } from "@huggingface/inference";
 
-const client = new InferenceClient(process.env.FIREWORKS_AI_API_KEY);
+const client = new InferenceClient(process.env.HF_TOKEN);
 
 let out = "";
 
diff --git a/packages/tasks-gen/snippets-fixtures/conversational-vlm-stream/python/huggingface_hub/0.fireworks-ai.py b/packages/tasks-gen/snippets-fixtures/conversational-vlm-stream/python/huggingface_hub/0.fireworks-ai.py
@@ -3,7 +3,7 @@
 
 client = InferenceClient(
     provider="fireworks-ai",
-    api_key=os.environ["FIREWORKS_AI_API_KEY"],
+    api_key=os.environ["HF_TOKEN"],
 )
 
 stream = client.chat.completions.create(
diff --git a/packages/tasks/src/hardware.ts b/packages/tasks/src/hardware.ts
@@ -38,6 +38,10 @@ export const DEFAULT_MEMORY_OPTIONS = [
 export const SKUS = {
 	GPU: {
 		NVIDIA: {
+			B200: {
+				tflops: 496.6,
+				memory: [192],
+			},
 			H200: {
 				tflops: 241.3,
 				memory: [141],
diff --git a/packages/tasks/src/model-libraries-snippets.ts b/packages/tasks/src/model-libraries-snippets.ts
@@ -1411,7 +1411,7 @@ export const voicecraft = (model: ModelData): string[] => [
 model = VoiceCraft.from_pretrained("${model.id}")`,
 ];
 
-export const vui = (model: ModelData): string[] => [
+export const vui = (): string[] => [
 	`# !pip install git+https://github.com/fluxions-ai/vui
 
 import torchaudio
diff --git a/packages/tasks/src/pipelines.ts b/packages/tasks/src/pipelines.ts
diff --git a/packages/tasks/src/tasks/text-generation/about.md b/packages/tasks/src/tasks/text-generation/about.md

-Original file line number
+Diff line change
 ```html
 <script type="module">
 -    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/[email protected].4/+esm';
 +    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/[email protected].6/+esm';
     import { createRepo, commit, deleteRepo, listFiles } from "https://cdn.jsdelivr.net/npm/@huggingface/[email protected]/+esm";
 </script>
 ```
Original file line number	Diff line number	Diff line change
`@@ -1,6 +1,6 @@`
`1`	`1`	`{`
`2`	`2`	`"name": "@huggingface/inference",`
`3`		`- "version": "4.0.4",`
	`3`	`+ "version": "4.0.6",`
`4`	`4`	`"packageManager": "[email protected]",`
`5`	`5`	`"license": "MIT",`
`6`	`6`	`"author": "Hugging Face and Tim Mikeladze <[email protected]>",`