update Nscale provider

Author: nbarr07

.github/workflows/test.yml

@@ -61,11 +61,13 @@ jobs:
             echo "SINCE=$(git merge-base origin/${{ github.event.pull_request.base.ref }} ${{ github.sha }})" >> $GITHUB_OUTPUT
           fi
 
+      - run: google-chrome --version
+
       - run: npm install -g corepack@latest && corepack enable
 
       - uses: actions/setup-node@v3
         with:
-          node-version: "20"
+          node-version: "22"
           cache: "pnpm"
           cache-dependency-path: "**/pnpm-lock.yaml"
       - run: |
@@ -105,7 +107,7 @@ jobs:
         run: |
           sleep 3
           pnpm i --filter root --filter inference... --filter hub... --filter tasks-gen --frozen-lockfile
-          pnpm --filter inference --filter hub --filter tasks publish --force --no-git-checks --registry http://localhost:4874/
+          pnpm --filter inference --filter hub --filter tasks --filter jinja publish --force --no-git-checks --registry http://localhost:4874/
 
       - name: E2E test - test yarn install
         working-directory: e2e/ts
@@ -136,7 +138,7 @@ jobs:
           deno-version: vx.x.x
       - name: E2E test - deno import from npm
         working-directory: e2e/deno
-        run: deno run --allow-net --allow-env=HF_TOKEN index.ts
+        run: deno run --allow-read --allow-net --allow-env=HF_TOKEN index.ts
         env:
           NPM_CONFIG_REGISTRY: http://localhost:4874/
           HF_TOKEN: ${{ secrets.HF_TOKEN }}

CODEOWNERS

@@ -1,6 +1,6 @@
 # Ownership for the Inference Package
 
-/packages/inference/ @julien-c @hanouticelina @SBrandeis @coyotte508
+/packages/inference/ @julien-c @hanouticelina @SBrandeis
 
 # Ownership for the Tasks Package
 

CONTRIBUTING.md

@@ -8,7 +8,7 @@ We use `pnpm` as our package manager. You need to use it, eg `pnpm install` inst
 
 If you want to format the whole codebase, you can do `pnpm -r format` at the root.
 
-Other than that, we avoid runtime dependencies unless they're strictly needed. For example, our only dependency is `hash-wasm`, and it's only in the browser context and when uploaded files are > 10MB.
+Other than that, we avoid runtime dependencies unless they're strictly needed. Even then, we prefer vendoring the code.
 
 ## Pull requests
 
@@ -18,7 +18,11 @@ It's not a hard requirement, but please consider using an icon from [Gitmoji](ht
 
 ## Tests
 
-If you want to run only specific tests, you can do `pnpm test -- -t "test name"`
+If you want to run only specific tests, you can do `pnpm test -- -t "test name"`.
+
+You can also do `npx vitest ./packages/hub/src/utils/XetBlob.spec.ts` to run a specific test file.
+
+Or `cd packages/hub && npx vitest --browser.name=chrome --browser.headless --config vitest-browser.config.mts ./src/utils/XetBlob.spec.ts` to run browser tests on a specific file
 
 ## Adding a package
 

README.md

@@ -10,7 +10,7 @@
 </p>
 
 ```ts
-// Programatically interact with the Hub
+// Programmatically interact with the Hub
 
 await createRepo({
   repo: { type: "model", name: "my-user/nlp-model" },
@@ -31,6 +31,7 @@ await uploadFile({
 
 await inference.chatCompletion({
   model: "meta-llama/Llama-3.1-8B-Instruct",
+  provider: "sambanova", // or together, fal-ai, replicate, cohere …
   messages: [
     {
       role: "user",
@@ -39,11 +40,11 @@ await inference.chatCompletion({
   ],
   max_tokens: 512,
   temperature: 0.5,
-  provider: "sambanova", // or together, fal-ai, replicate, cohere …
 });
 
 await inference.textToImage({
   model: "black-forest-labs/FLUX.1-dev",
+  provider: "replicate",
   inputs: "a picture of a green bird",
 });
 
@@ -54,7 +55,7 @@ await inference.textToImage({
 
 This is a collection of JS libraries to interact with the Hugging Face API, with TS types included.
 
-- [@huggingface/inference](packages/inference/README.md): Use HF Inference API (serverless), Inference Endpoints (dedicated) and third-party Inference Providers to make calls to 100,000+ Machine Learning models
+- [@huggingface/inference](packages/inference/README.md): Use HF Inference API (serverless), Inference Endpoints (dedicated) and all supported Inference Providers to make calls to 100,000+ Machine Learning models
 - [@huggingface/hub](packages/hub/README.md): Interact with huggingface.co to create or delete repos and commit / download files
 - [@huggingface/agents](packages/agents/README.md): Interact with HF models through a natural language interface
 - [@huggingface/gguf](packages/gguf/README.md): A GGUF parser that works on remotely hosted files.
@@ -84,7 +85,7 @@ npm install @huggingface/agents
 Then import the libraries in your code:
 
 ```ts
-import { HfInference } from "@huggingface/inference";
+import { InferenceClient } from "@huggingface/inference";
 import { HfAgent } from "@huggingface/agents";
 import { createRepo, commit, deleteRepo, listFiles } from "@huggingface/hub";
 import type { RepoId } from "@huggingface/hub";
@@ -96,21 +97,21 @@ You can run our packages with vanilla JS, without any bundler, by using a CDN or
 
 ```html
 <script type="module">
-    import { HfInference } from 'https://cdn.jsdelivr.net/npm/@huggingface/inference@3.5.1/+esm';
-    import { createRepo, commit, deleteRepo, listFiles } from "https://cdn.jsdelivr.net/npm/@huggingface/hub@1.0.2/+esm";
+    import { InferenceClient } from 'https://cdn.jsdelivr.net/npm/@huggingface/inference@3.7.0/+esm';
+    import { createRepo, commit, deleteRepo, listFiles } from "https://cdn.jsdelivr.net/npm/@huggingface/hub@1.1.2/+esm";
 </script>
 ```
 
 ### Deno
 
 ```ts
 // esm.sh
-import { HfInference } from "https://esm.sh/@huggingface/inference"
+import { InferenceClient } from "https://esm.sh/@huggingface/inference"
 import { HfAgent } from "https://esm.sh/@huggingface/agents";
 
 import { createRepo, commit, deleteRepo, listFiles } from "https://esm.sh/@huggingface/hub"
 // or npm:
-import { HfInference } from "npm:@huggingface/inference"
+import { InferenceClient } from "npm:@huggingface/inference"
 import { HfAgent } from "npm:@huggingface/agents";
 
 import { createRepo, commit, deleteRepo, listFiles } from "npm:@huggingface/hub"
@@ -123,11 +124,11 @@ Get your HF access token in your [account settings](https://huggingface.co/setti
 ### @huggingface/inference examples
 
 ```ts
-import { HfInference } from "@huggingface/inference";
+import { InferenceClient } from "@huggingface/inference";
 
 const HF_TOKEN = "hf_...";
 
-const inference = new HfInference(HF_TOKEN);
+const inference = new InferenceClient(HF_TOKEN);
 
 // Chat completion API
 const out = await inference.chatCompletion({
@@ -179,7 +180,7 @@ await inference.imageToText({
 
 // Using your own dedicated inference endpoint: https://hf.co/docs/inference-endpoints/
 const gpt2 = inference.endpoint('https://xyz.eu-west-1.aws.endpoints.huggingface.cloud/gpt2');
-const { generated_text } = await gpt2.textGeneration({inputs: 'The answer to the universe is'});
+const { generated_text } = await gpt2.textGeneration({ inputs: 'The answer to the universe is' });
 
 // Chat Completion
 const llamaEndpoint = inference.endpoint(

e2e/deno/index.ts

@@ -1,4 +1,4 @@
-import { HfInference } from "npm:@huggingface/inference@*";
+import { InferenceClient } from "npm:@huggingface/inference@*";
 import { whoAmI, listFiles } from "npm:@huggingface/hub@*";
 
 const info = await whoAmI({ credentials: { accessToken: "hf_hub.js" }, hubUrl: "https://hub-ci.huggingface.co" });
@@ -10,7 +10,7 @@ for await (const file of listFiles({ repo: "gpt2" })) {
 
 const token = Deno.env.get("HF_TOKEN");
 if (token) {
-	const hf = new HfInference(token);
+	const hf = new InferenceClient(token);
 
 	const tokenInfo = await whoAmI({ credentials: { accessToken: token } });
 	console.log(tokenInfo);

e2e/svelte/src/routes/+page.svelte

@@ -1,8 +1,8 @@
 <script>
 	import { whoAmI, listFiles } from "@huggingface/hub";
-	import { HfInference } from "@huggingface/inference";
+	import { InferenceClient } from "@huggingface/inference";
 
-	const hf = new HfInference();
+	const hf = new InferenceClient();
 
 	const test = async () => {
 		const info = await whoAmI({ credentials: { accessToken: "hf_hub.js" }, hubUrl: "https://hub-ci.huggingface.co" });

e2e/ts/src/index.ts

@@ -1,9 +1,9 @@
-import { HfInference } from "@huggingface/inference";
+import { InferenceClient } from "@huggingface/inference";
 import { whoAmI } from "@huggingface/hub";
 
 const hfToken = process.env.token;
 
-const hf = new HfInference(hfToken);
+const hf = new InferenceClient(hfToken);
 
 (async () => {
 	const info = await whoAmI({ credentials: { accessToken: "hf_hub.js" }, hubUrl: "https://hub-ci.huggingface.co" });

packages/agents/README.md

@@ -58,7 +58,7 @@ const agent = new HfAgent(
 
 
 #### From your own endpoints
-You can also specify your own endpoint, as long as it implements the same API, for exemple using [text generation inference](https://github.com/huggingface/text-generation-inference) and [Inference Endpoints](https://huggingface.co/inference-endpoints).
+You can also specify your own endpoint, as long as it implements the same API, for example using [text generation inference](https://github.com/huggingface/text-generation-inference) and [Inference Endpoints](https://huggingface.co/inference-endpoints).
 
 ```ts
 import { HfAgent, LLMFromEndpoint } from "@huggingface/agents";

packages/agents/package.json

@@ -56,6 +56,6 @@
 		"@types/node": "^18.13.0"
 	},
 	"dependencies": {
-		"@huggingface/inference": "^2.6.1"
+		"@huggingface/inference": "workspace:^"
 	}
 }

packages/agents/pnpm-lock.yaml

@@ -1,4 +1,4 @@
-import { HfInference } from "@huggingface/inference";
+import { InferenceClient } from "@huggingface/inference";
 import type { Data, Tool } from "../types";
 
 // this function passes the tools & files to the context before calling eval
@@ -17,7 +17,7 @@ export async function evalBuilder(
 
 		// add tools to context
 		for (const tool of tools) {
-			const toolCall = (input: Promise<Data>) => tool.call?.(input, new HfInference(accessToken ?? ""));
+			const toolCall = (input: Promise<Data>) => tool.call?.(input, new InferenceClient(accessToken ?? ""));
 			// @ts-expect-error adding to the scope
 			globalThis[tool.name] = toolCall;
 		}

packages/agents/src/lib/evalBuilder.ts

@@ -1,8 +1,8 @@
 import type { LLM } from "../types";
-import { HfInference } from "@huggingface/inference";
+import { InferenceClient } from "@huggingface/inference";
 
 export function LLMFromHub(accessToken?: string, model?: string): LLM {
-	const inference = new HfInference(accessToken);
+	const inference = new InferenceClient(accessToken);
 
 	return async (prompt: string): Promise<string> => {
 		const formattedPrompt = "<|user|>" + prompt + "<|end|><|assistant|>";
@@ -20,7 +20,7 @@ export function LLMFromHub(accessToken?: string, model?: string): LLM {
 }
 
 export function LLMFromEndpoint(accessToken: string, endpoint: string): LLM {
-	const inference = new HfInference(accessToken).endpoint(endpoint);
+	const inference = new InferenceClient(accessToken).endpoint(endpoint);
 	return async (prompt: string): Promise<string> => {
 		const formattedPrompt = "<|user|>" + prompt + "<|end|><|assistant|>";
 

packages/agents/src/llms/LLMHF.ts

@@ -15,9 +15,12 @@ export const imageToTextTool: Tool = {
 		if (typeof data === "string") throw "Input must be a blob.";
 
 		return (
-			await inference.imageToText({
-				data,
-			})
-		).generated_text;
+			// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+			(
+				await inference.imageToText({
+					data,
+				})
+			).generated_text!
+		);
 	},
 };

packages/agents/src/tools/imageToText.ts

@@ -1,12 +1,12 @@
-import type { HfInference } from "@huggingface/inference";
+import type { InferenceClient } from "@huggingface/inference";
 
 export type Data = string | Blob | ArrayBuffer;
 
 export interface Tool {
 	name: string;
 	description: string;
 	examples: Array<Example>;
-	call?: (input: Promise<Data>, inference: HfInference) => Promise<Data>;
+	call?: (input: Promise<Data>, inference: InferenceClient) => Promise<Data>;
 }
 
 export interface Example {

packages/agents/src/types.d.ts

@@ -1,7 +1,7 @@
 import { describe, expect, it } from "vitest";
 import { HfAgent, defaultTools, LLMFromHub, LLMFromEndpoint } from "../src";
 import type { Data } from "../src/types";
-import type { HfInference } from "@huggingface/inference";
+import type { InferenceClient } from "@huggingface/inference";
 
 const env = import.meta.env;
 if (!env.HF_TOKEN) {
@@ -33,7 +33,7 @@ describe("HfAgent", () => {
 				},
 			],
 			// eslint-disable-next-line @typescript-eslint/no-unused-vars
-			call: async (input: Promise<Data>, inference: HfInference): Promise<Data> => {
+			call: async (input: Promise<Data>, inference: InferenceClient): Promise<Data> => {
 				const data = await input;
 				if (typeof data !== "string") {
 					throw new Error("Input must be a string");

packages/agents/test/HfAgent.spec.ts

@@ -1,6 +1,7 @@
 #!/usr/bin/env node
 
-import { GGMLQuantizationType, gguf, ggufAllShards, GGUFParseOutput } from ".";
+import type { GGUFParseOutput } from ".";
+import { GGMLQuantizationType, ggufAllShards } from ".";
 import { GGML_QUANT_SIZES } from "./quant-descriptions";
 
 interface PrintColumnHeader {

packages/gguf/src/cli.ts

@@ -1,2 +1,3 @@
 dist
 sha256.js
+src/vendor

packages/hub/.eslintignore

@@ -174,3 +174,4 @@ Under the hood, `@huggingface/hub` uses a lazy blob implementation to load the f
 ## Dependencies
 
 - `@huggingface/tasks` : Typings only
+- `@huggingface/lz4` : URL join utility

packages/hub/README.md

@@ -1,7 +1,7 @@
 {
 	"name": "@huggingface/hub",
 	"packageManager": "[email protected]",
-	"version": "1.0.2",
+	"version": "1.1.2",
 	"description": "Utilities to interact with the Hugging Face hub",
 	"repository": "https://github.com/huggingface/huggingface.js.git",
 	"publishConfig": {

packages/hub/package.json

@@ -22,3 +22,4 @@ export { HubApiError, InvalidApiResponseFormatError } from "./error";
  * Only exported for E2Es convenience
  */
 export { sha256 as __internal_sha256 } from "./utils/sha256";
+export { XetBlob as __internal_XetBlob } from "./utils/XetBlob";

packages/hub/src/index.ts

@@ -7,6 +7,7 @@ import type { Stats } from "node:fs";
 import { getHFHubCachePath, getRepoFolderName } from "./cache-management";
 import { toRepoId } from "../utils/toRepoId";
 import { downloadFileToCacheDir } from "./download-file-to-cache-dir";
+import { createSymlink } from "../utils/symlink";
 
 vi.mock("node:fs/promises", () => ({
 	writeFile: vi.fn(),
@@ -21,6 +22,10 @@ vi.mock("./paths-info", () => ({
 	pathsInfo: vi.fn(),
 }));
 
+vi.mock("../utils/symlink", () => ({
+	createSymlink: vi.fn(),
+}));
+
 const DUMMY_REPO: RepoId = {
 	name: "hello-world",
 	type: "model",
@@ -196,7 +201,7 @@ describe("downloadFileToCacheDir", () => {
 		expect(vi.mocked(lstat).mock.calls[0][0]).toBe(expectedBlob);
 
 		// symlink should have been created
-		expect(symlink).toHaveBeenCalledOnce();
+		expect(createSymlink).toHaveBeenCalledOnce();
 		// no download done
 		expect(fetchMock).not.toHaveBeenCalled();
 
@@ -283,6 +288,6 @@ describe("downloadFileToCacheDir", () => {
 		// 2. should rename the incomplete to the blob expected name
 		expect(rename).toHaveBeenCalledWith(incomplete, expectedBlob);
 		// 3. should create symlink pointing to blob
-		expect(symlink).toHaveBeenCalledWith(expectedBlob, expectPointer);
+		expect(createSymlink).toHaveBeenCalledWith({ sourcePath: expectedBlob, finalPath: expectPointer });
 	});
 });