feat: support zai as inference provider (#1766)

Hi 🤗 team, 
We’d like to apply to register [ZAI](https://z.ai/) as inference
provider. Thanks

One more question: when we tried to register the model mapping with the
endpoint `POST api/partners/zai/models`, we received a 404 response.
https://huggingface.co/docs/inference-providers/register-as-a-provider#3-model-mapping-api
Does this action only work after the PR has been merged?

---------

Co-authored-by: gongchao <[email protected]>
Co-authored-by: célina <[email protected]>
Co-authored-by: Simon Brandeis <[email protected]>
Co-authored-by: SBrandeis <[email protected]>

Author: 4 people

packages/inference/README.md

@@ -65,6 +65,7 @@ Currently, we support the following providers:
 - [Cohere](https://cohere.com)
 - [Cerebras](https://cerebras.ai/)
 - [Groq](https://groq.com)
+- [Z.ai](https://z.ai/)
 
 To send requests to a third-party provider, you have to pass the `provider` parameter to the inference function. The default value of the `provider` parameter is "auto", which will select the first of the providers available for the model, sorted by your preferred order in https://hf.co/settings/inference-providers.
 
@@ -100,6 +101,7 @@ Only a subset of models are supported when requesting third-party providers. You
 - [Cerebras supported models](https://huggingface.co/api/partners/cerebras/models)
 - [Groq supported models](https://console.groq.com/docs/models)
 - [Novita AI supported models](https://huggingface.co/api/partners/novita/models)
+- [Z.ai supported models](https://huggingface.co/api/partners/zai-org/models)
 
 ❗**Important note:** To be compatible, the third-party API must adhere to the "standard" shape API we expect on HF model pages for each pipeline task type.
 This is not an issue for LLMs as everyone converged on the OpenAI API anyways, but can be more tricky for other tasks like "text-to-image" or "automatic-speech-recognition" where there exists no standard API. Let us know if any help is needed or if we can make things easier for you!

packages/inference/src/lib/getProviderHelper.ts

@@ -50,6 +50,7 @@ import * as Replicate from "../providers/replicate.js";
 import * as Sambanova from "../providers/sambanova.js";
 import * as Scaleway from "../providers/scaleway.js";
 import * as Together from "../providers/together.js";
+import * as Zai from "../providers/zai-org.js";
 import type { InferenceProvider, InferenceProviderOrPolicy, InferenceTask } from "../types.js";
 import { InferenceClientInputError } from "../errors.js";
 
@@ -164,6 +165,9 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 		conversational: new Together.TogetherConversationalTask(),
 		"text-generation": new Together.TogetherTextGenerationTask(),
 	},
+	"zai-org": {
+		conversational: new Zai.ZaiConversationalTask(),
+	},
 };
 
 /**

packages/inference/src/providers/consts.ts

@@ -37,4 +37,5 @@ export const HARDCODED_MODEL_INFERENCE_MAPPING: Record<
 	sambanova: {},
 	scaleway: {},
 	together: {},
+	"zai-org": {},
 };

packages/inference/src/providers/zai-org.ts

@@ -0,0 +1,37 @@
+/**
+ * See the registered mapping of HF model ID => ZAI model ID here:
+ *
+ * https://huggingface.co/api/partners/zai-org/models
+ *
+ * This is a publicly available mapping.
+ *
+ * If you want to try to run inference for a new model locally before it's registered on huggingface.co,
+ * you can add it to the dictionary "HARDCODED_MODEL_ID_MAPPING" in consts.ts, for dev purposes.
+ *
+ * - If you work at zai and want to update this mapping, please use the model mapping API we provide on huggingface.co
+ * - If you're a community member and want to add a new supported HF model to zai, please open an issue on the present repo
+ * and we will tag zai team members.
+ *
+ * Thanks!
+ */
+import { BaseConversationalTask } from "./providerHelper.js";
+import type { HeaderParams } from "../types.js";
+
+const ZAI_API_BASE_URL = "https://api.z.ai";
+
+export class ZaiConversationalTask extends BaseConversationalTask {
+	constructor() {
+		super("zai-org", ZAI_API_BASE_URL);
+	}
+
+	override prepareHeaders(params: HeaderParams, binary: boolean): Record<string, string> {
+		const headers = super.prepareHeaders(params, binary);
+		headers["x-source-channel"] = "hugging_face";
+		headers["accept-language"] = "en-US,en";
+		return headers;
+	}
+
+	override makeRoute(): string {
+		return "/api/paas/v4/chat/completions";
+	}
+}

packages/inference/src/types.ts

@@ -64,6 +64,7 @@ export const INFERENCE_PROVIDERS = [
 	"sambanova",
 	"scaleway",
 	"together",
+	"zai-org",
 ] as const;
 
 export const PROVIDERS_OR_POLICIES = [...INFERENCE_PROVIDERS, "auto"] as const;

packages/inference/test/InferenceClient.spec.ts

@@ -2134,6 +2134,58 @@ describe.skip("InferenceClient", () => {
 		},
 		TIMEOUT
 	);
+	describe.concurrent(
+		"ZAI",
+		() => {
+			const client = new InferenceClient(env.HF_ZAI_KEY ?? "dummy");
+
+			HARDCODED_MODEL_INFERENCE_MAPPING["zai-org"] = {
+				"zai-org/GLM-4.5": {
+					provider: "zai-org",
+					hfModelId: "zai-org/GLM-4.5",
+					providerId: "glm-4.5",
+					status: "live",
+					task: "conversational",
+				},
+			};
+
+			it("chatCompletion", async () => {
+				const res = await client.chatCompletion({
+					model: "zai-org/GLM-4.5",
+					provider: "zai-org",
+					messages: [{ role: "user", content: "Complete this sentence with words, one plus one is equal " }],
+				});
+				if (res.choices && res.choices.length > 0) {
+					const completion = res.choices[0].message?.content;
+					expect(completion).toContain("two");
+				}
+			});
+
+			it("chatCompletion stream", async () => {
+				const stream = client.chatCompletionStream({
+					model: "zai-org/GLM-4.5",
+					provider: "zai-org",
+					messages: [{ role: "user", content: "Say 'this is a test'" }],
+					stream: true,
+				}) as AsyncGenerator<ChatCompletionStreamOutput>;
+
+				let fullResponse = "";
+				for await (const chunk of stream) {
+					if (chunk.choices && chunk.choices.length > 0) {
+						const content = chunk.choices[0].delta?.content;
+						if (content) {
+							fullResponse += content;
+						}
+					}
+				}
+
+				// Verify we got a meaningful response
+				expect(fullResponse).toBeTruthy();
+				expect(fullResponse.length).toBeGreaterThan(0);
+			});
+		},
+		TIMEOUT
+	);
 	describe.concurrent(
 		"OVHcloud",
 		() => {