Add OVHcloud as an inference provider

packages/inference/README.md

@@ -58,6 +58,7 @@ Currently, we support the following providers:
 - [Blackforestlabs](https://blackforestlabs.ai)
 - [Cohere](https://cohere.com)
 - [Cerebras](https://cerebras.ai/)
+- [OVHcloud](https://endpoints.ai.cloud.ovh.net/)
 
 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.
 ```ts
@@ -84,6 +85,7 @@ Only a subset of models are supported when requesting third-party providers. You
 - [Together supported models](https://huggingface.co/api/partners/together/models)
 - [Cohere supported models](https://huggingface.co/api/partners/cohere/models)
 - [Cerebras supported models](https://huggingface.co/api/partners/cerebras/models)
+- [OVHcloud supported models](https://huggingface.co/api/partners/ovhcloud/models)
 - [HF Inference API (serverless)](https://huggingface.co/models?inference=warm&sort=trending)
 
 ❗**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.

packages/inference/src/lib/getProviderHelper.ts

@@ -9,6 +9,7 @@ import * as Hyperbolic from "../providers/hyperbolic";
 import * as Nebius from "../providers/nebius";
 import * as Novita from "../providers/novita";
 import * as OpenAI from "../providers/openai";
+import * as OvhCloud from "../providers/ovhcloud";
 import type {
 	AudioClassificationTaskHelper,
 	AudioToAudioTaskHelper,
@@ -112,6 +113,10 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 	openai: {
 		conversational: new OpenAI.OpenAIConversationalTask(),
 	},
+	ovhcloud: {
+		conversational: new OvhCloud.OvhCloudConversationalTask(),
+		"text-generation": new OvhCloud.OvhCloudTextGenerationTask(),
+	},
 	replicate: {
 		"text-to-image": new Replicate.ReplicateTextToImageTask(),
 		"text-to-speech": new Replicate.ReplicateTextToSpeechTask(),

packages/inference/src/providers/consts.ts

@@ -29,6 +29,7 @@ export const HARDCODED_MODEL_INFERENCE_MAPPING: Record<
 	nebius: {},
 	novita: {},
 	openai: {},
+	ovhcloud: {},
 	replicate: {},
 	sambanova: {},
 	together: {},

packages/inference/src/providers/ovhcloud.ts

@@ -0,0 +1,89 @@
+/**
+ * See the registered mapping of HF model ID => OVHcloud model ID here:
+ *
+ * https://huggingface.co/api/partners/ovhcloud/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 OVHcloud 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 OVHcloud, please open an issue on the present repo
+ * and we will tag OVHcloud team members.
+ *
+ * Thanks!
+ */
+
+import { BaseConversationalTask, BaseTextGenerationTask } from "./providerHelper";
+import type {
+	ChatCompletionOutput,
+	TextGenerationOutput,
+	TextGenerationOutputFinishReason,
+} from "@huggingface/tasks";
+import { InferenceOutputError } from "../lib/InferenceOutputError";
+import type { BodyParams } from "../types";
+import { omit } from "../utils/omit";
+
+const OVHCLOUD_API_BASE_URL = "https://oai.endpoints.kepler.ai.cloud.ovh.net";
+
+function prepareBaseOvhCloudPayload(params: BodyParams): Record<string, unknown> {
+	return {
+		model: params.model,
+		...omit(params.args, ["inputs", "parameters"]),
+		...(params.args.parameters
+			? {
+				max_tokens: (params.args.parameters as Record<string, unknown>).max_new_tokens,
+				...omit(params.args.parameters as Record<string, unknown>, "max_new_tokens"),
+			}
+			: undefined),
+		prompt: params.args.inputs,
+	};
+}
+
+interface OvhCloudTextCompletionOutput extends Omit<ChatCompletionOutput, "choices"> {
+	choices: Array<{
+		text: string;
+		finish_reason: TextGenerationOutputFinishReason;
+		logprobs: unknown;
+		index: number;
+	}>;
+}
+
+export class OvhCloudConversationalTask extends BaseConversationalTask {
+	constructor() {
+		super("ovhcloud", OVHCLOUD_API_BASE_URL);
+	}
+
+	override preparePayload(params: BodyParams): Record<string, unknown> {
+		return prepareBaseOvhCloudPayload(params);
+	}
+}
+
+export class OvhCloudTextGenerationTask extends BaseTextGenerationTask {
+	constructor() {
+		super("ovhcloud", OVHCLOUD_API_BASE_URL);
+	}
+
+	override preparePayload(params: BodyParams): Record<string, unknown> {
+		const payload = prepareBaseOvhCloudPayload(params);
+		payload.prompt = params.args.inputs;
+		return payload;
+	}
+
+	override async getResponse(response: OvhCloudTextCompletionOutput): Promise<TextGenerationOutput> {
+		if (
+			typeof response === "object" &&
+			"choices" in response &&
+			Array.isArray(response?.choices) &&
+			typeof response?.model === "string"
+		) {
+			const completion = response.choices[0];
+			return {
+				generated_text: completion.text,
+			};
+		}
+		throw new InferenceOutputError("Expected OVHcloud text generation response format");
+	}
+
+}

packages/inference/src/types.ts

@@ -48,6 +48,7 @@ export const INFERENCE_PROVIDERS = [
 	"nebius",
 	"novita",
 	"openai",
+	"ovhcloud",
 	"replicate",
 	"sambanova",
 	"together",

packages/inference/test/InferenceClient.spec.ts

@@ -1691,4 +1691,111 @@ describe.concurrent("InferenceClient", () => {
 		},
 		TIMEOUT
 	);
+	describe.concurrent(
+		"OVHcloud",
+		() => {
+			const client = new HfInference(env.HF_OVHCLOUD_KEY ?? "dummy");
+
+			HARDCODED_MODEL_INFERENCE_MAPPING["ovhcloud"] = {
+				"meta-llama/llama-3.1-8b-instruct": {
+					hfModelId: "meta-llama/llama-3.1-8b-instruct",
+					providerId: "Llama-3.1-8B-Instruct",
+					status: "live",
+					task: "conversational",
+				},
+			};
+
+			it("chatCompletion", async () => {
+				const res = await client.chatCompletion({
+					model: "meta-llama/llama-3.1-8b-instruct",
+					provider: "ovhcloud",
+					messages: [{ role: "user", content: "A, B, C, " }],
+					parameters: {
+						seed: 42,
+						temperature: 0,
+						top_p: 0.01,
+						max_new_tokens: 1,
+					},
+				});
+				expect(res.choices && res.choices.length > 0);
+				const completion = res.choices[0].message?.content;
+				expect(completion).toContain("D");
+			});
+
+			it("chatCompletion stream", async () => {
+				const stream = client.chatCompletionStream({
+					model: "meta-llama/llama-3.1-8b-instruct",
+					provider: "ovhcloud",
+					messages: [{ role: "user", content: "A, B, C, " }],
+					stream: true,
+					parameters: {
+						seed: 42,
+						temperature: 0,
+						top_p: 0.01,
+						max_new_tokens: 1,
+					},
+				}) 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).toContain("D");
+			});
+
+			it("textGeneration", async () => {
+				const res = await client.textGeneration({
+					model: "meta-llama/llama-3.1-8b-instruct",
+					provider: "ovhcloud",
+					inputs: "A B C ",
+					parameters: {
+						seed: 42,
+						temperature: 0,
+						top_p: 0.01,
+						max_new_tokens: 1,
+					},
+				});
+				expect(res.generated_text.length > 0);
+				expect(res.generated_text).toContain("D");
+			});
+
+			it("textGeneration stream", async () => {
+				const stream = client.textGenerationStream({
+					model: "meta-llama/llama-3.1-8b-instruct",
+					provider: "ovhcloud",
+					inputs: "A B C ",
+					stream: true,
+					parameters: {
+						seed: 42,
+						temperature: 0,
+						top_p: 0.01,
+						max_new_tokens: 1,
+					},
+				}) as AsyncGenerator<ChatCompletionStreamOutput>;
+
+				let fullResponse = "";
+				for await (const chunk of stream) {
+					if (chunk.choices && chunk.choices.length > 0) {
+						const content = chunk.choices[0].text;
+						if (content) {
+							fullResponse += content;
+						}
+					}
+				}
+
+				// Verify we got a meaningful response
+				expect(fullResponse).toBeTruthy();
+				expect(fullResponse).toContain("D");
+			});
+		},
+		TIMEOUT
+	);
 });

packages/tasks/src/inference-providers.ts

@@ -7,6 +7,7 @@ const INFERENCE_PROVIDERS = [
 	"fireworks-ai",
 	"hf-inference",
 	"hyperbolic",
+	"ovhcloud",
 	"replicate",
 	"sambanova",
 	"together",