@huggingface/inference - npm Package Compare versions

Comparing version 3.4.1 to 3.5.0

dist/src/index.d.ts

@@ -5,2 +5,4 @@ export { HfInference, HfInferenceEndpoint } from "./HfInference";
 export * from "./tasks";
+import * as snippets from "./snippets/index.js";
+export { snippets };
 //# sourceMappingURL=index.d.ts.map

dist/src/types.d.ts

@@ -25,3 +25,3 @@ import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
 export type InferenceTask = Exclude<PipelineType, "other">;
-export declare const INFERENCE_PROVIDERS: readonly ["black-forest-labs", "cohere", "fal-ai", "fireworks-ai", "hf-inference", "hyperbolic", "nebius", "novita", "replicate", "sambanova", "together"];
+export declare const INFERENCE_PROVIDERS: readonly ["black-forest-labs", "cerebras", "cohere", "fal-ai", "fireworks-ai", "hf-inference", "hyperbolic", "nebius", "novita", "openai", "replicate", "sambanova", "together"];
 export type InferenceProvider = (typeof INFERENCE_PROVIDERS)[number];
@@ -77,2 +77,3 @@ export interface BaseArgs {
     makeUrl: (params: UrlParams) => string;
+    clientSideRoutingOnly?: boolean;
 }
@@ -79,0 +80,0 @@ export interface HeaderParams {

package.json

 {
   "name": "@huggingface/inference",
-  "version": "3.4.1",
+  "version": "3.5.0",
   "packageManager": "pnpm@8.10.5",
@@ -42,3 +42,3 @@ "license": "MIT",
   "dependencies": {
-    "@huggingface/tasks": "^0.16.5"
+    "@huggingface/tasks": "^0.17.0"
   },
@@ -45,0 +45,0 @@ "devDependencies": {

README.md

@@ -60,2 +60,3 @@ # 🤗 Hugging Face Inference
 - [Cohere](https://cohere.com)
+- [Cerebras](https://cerebras.ai/)
 
@@ -86,2 +87,3 @@ To send requests to a third-party provider, you have to pass the `provider` parameter to the inference function. Make sure your request is authenticated with an access token.
 - [Cohere supported models](https://huggingface.co/api/partners/cohere/models)
+- [Cerebras supported models](https://huggingface.co/api/partners/cerebras/models)
 - [HF Inference API (serverless)](https://huggingface.co/models?inference=warm&sort=trending)
@@ -88,0 +90,0 @@

src/index.ts

@@ -5,1 +5,4 @@ export { HfInference, HfInferenceEndpoint } from "./HfInference";
 export * from "./tasks";
+
+import * as snippets from "./snippets/index.js";
+export { snippets };

src/lib/makeRequestOptions.ts

 import { HF_HUB_URL, HF_ROUTER_URL } from "../config";
 import { BLACK_FOREST_LABS_CONFIG } from "../providers/black-forest-labs";
+import { CEREBRAS_CONFIG } from "../providers/cerebras";
 import { COHERE_CONFIG } from "../providers/cohere";
@@ -13,2 +14,3 @@ import { FAL_AI_CONFIG } from "../providers/fal-ai";
 import { TOGETHER_CONFIG } from "../providers/together";
+import { OPENAI_CONFIG } from "../providers/openai";
 import type { InferenceProvider, InferenceTask, Options, ProviderConfig, RequestArgs } from "../types";
@@ -32,2 +34,3 @@ import { isUrl } from "./isUrl";
 	"black-forest-labs": BLACK_FOREST_LABS_CONFIG,
+	cerebras: CEREBRAS_CONFIG,
 	cohere: COHERE_CONFIG,
@@ -38,2 +41,3 @@ "fal-ai": FAL_AI_CONFIG,
 	hyperbolic: HYPERBOLIC_CONFIG,
+	openai: OPENAI_CONFIG,
 	nebius: NEBIUS_CONFIG,
@@ -78,18 +82,34 @@ novita: NOVITA_CONFIG,
 	}
+	if (providerConfig.clientSideRoutingOnly && !maybeModel) {
+		throw new Error(`Provider ${provider} requires a model ID to be passed directly.`);
+	}
 	// eslint-disable-next-line @typescript-eslint/no-non-null-assertion
 	const hfModel = maybeModel ?? (await loadDefaultModel(task!));
-	const model = await getProviderModelId({ model: hfModel, provider }, args, {
-		task,
-		chatCompletion,
-		fetch: options?.fetch,
-	});
+	const model = providerConfig.clientSideRoutingOnly
+		? // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+		  removeProviderPrefix(maybeModel!, provider)
+		: // For closed-models API providers, one needs to pass the model ID directly (e.g. "gpt-3.5-turbo")
+		  await getProviderModelId({ model: hfModel, provider }, args, {
+				task,
+				chatCompletion,
+				fetch: options?.fetch,
+		  });
 
-	/// If accessToken is passed, it should take precedence over includeCredentials
-	const authMethod = accessToken
-		? accessToken.startsWith("hf_")
-			? "hf-token"
-			: "provider-key"
-		: includeCredentials === "include"
-		  ? "credentials-include"
-		  : "none";
+	const authMethod = (() => {
+		if (providerConfig.clientSideRoutingOnly) {
+			// Closed-source providers require an accessToken (cannot be routed).
+			if (accessToken && accessToken.startsWith("hf_")) {
+				throw new Error(`Provider ${provider} is closed-source and does not support HF tokens.`);
+			}
+			return "provider-key";
+		}
+		if (accessToken) {
+			return accessToken.startsWith("hf_") ? "hf-token" : "provider-key";
+		}
+		if (includeCredentials === "include") {
+			// If accessToken is passed, it should take precedence over includeCredentials
+			return "credentials-include";
+		}
+		return "none";
+	})();
 
@@ -183,1 +203,8 @@ // Make URL
 }
+
+function removeProviderPrefix(model: string, provider: string): string {
+	if (!model.startsWith(`${provider}/`)) {
+		throw new Error(`Models from ${provider} must be prefixed by "${provider}/". Got "${model}".`);
+	}
+	return model.slice(provider.length + 1);
+}

src/providers/consts.ts

@@ -20,2 +20,3 @@ import type { InferenceProvider } from "../types";
 	"black-forest-labs": {},
+	cerebras: {},
 	cohere: {},
@@ -28,2 +29,3 @@ "fal-ai": {},
 	novita: {},
+	openai: {},
 	replicate: {},
@@ -30,0 +32,0 @@ sambanova: {},

src/types.ts

@@ -33,2 +33,3 @@ import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
 	"black-forest-labs",
+	"cerebras",
 	"cohere",
@@ -41,2 +42,3 @@ "fal-ai",
 	"novita",
+	"openai",
 	"replicate",
@@ -101,2 +103,3 @@ "sambanova",
 	makeUrl: (params: UrlParams) => string;
+	clientSideRoutingOnly?: boolean;
 }
@@ -103,0 +106,0 @@

No alert changes

Improved metrics

Total package byte prevSize

414467

34.57%

Number of package files

232

8.41%

Lines of code

10029

45.54%

Number of lines in readme file

678

0.3%

Dependency changes

• + Added

  @huggingface/tasks@0.17.4
  (transitive)

• - Removed

  @huggingface/tasks@0.16.7
  (transitive)

• Updated

  @huggingface/tasks@^0.17.0