Add CentML as an inference provider

packages/inference/README.md

@@ -60,6 +60,7 @@ Currently, we support the following providers:
 - [Blackforestlabs](https://blackforestlabs.ai)
 - [Cohere](https://cohere.com)
 - [Cerebras](https://cerebras.ai/)
+- [CentML](https://centml.ai)
 - [Groq](https://groq.com)
 
 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.
@@ -89,6 +90,7 @@ 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)
+- [CentML supported models](https://huggingface.co/api/partners/centml/models)
 - [Groq supported models](https://console.groq.com/docs/models)
 - [HF Inference API (serverless)](https://huggingface.co/models?inference=warm&sort=trending)
 

packages/inference/src/lib/getProviderHelper.ts

@@ -1,5 +1,6 @@
 import * as BlackForestLabs from "../providers/black-forest-labs";
 import * as Cerebras from "../providers/cerebras";
+import * as CentML from "../providers/centml";
 import * as Cohere from "../providers/cohere";
 import * as FalAI from "../providers/fal-ai";
 import * as FeatherlessAI from "../providers/featherless-ai";
@@ -55,6 +56,10 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 	cerebras: {
 		conversational: new Cerebras.CerebrasConversationalTask(),
 	},
+	centml: {
+		conversational: new CentML.CentMLConversationalTask(),
+		"text-generation": new CentML.CentMLTextGenerationTask(),
+	},
 	cohere: {
 		conversational: new Cohere.CohereConversationalTask(),
 	},

packages/inference/src/providers/centml.ts

@@ -0,0 +1,71 @@
+/**
+ * CentML provider implementation for serverless inference.
+ * This provider supports chat completions and text generation through CentML's serverless endpoints.
+ */
+import type { ChatCompletionOutput, TextGenerationOutput } from "@huggingface/tasks";
+import { InferenceOutputError } from "../lib/InferenceOutputError";
+import type { BodyParams } from "../types";
+import { BaseConversationalTask, BaseTextGenerationTask } from "./providerHelper";
+
+const CENTML_API_BASE_URL = "https://api.centml.com";
+
+export class CentMLConversationalTask extends BaseConversationalTask {
+	constructor() {
+		super("centml", CENTML_API_BASE_URL);
+	}
+
+	override makeRoute(): string {
+		return "openai/v1/chat/completions";
+	}
+
+	override preparePayload(params: BodyParams): Record<string, unknown> {
+		const { args, model } = params;
+		return {
+			...args,
+			model,
+			api_key: args.accessToken,
+		};
+	}
+
+	override async getResponse(response: ChatCompletionOutput): Promise<ChatCompletionOutput> {
+		if (
+			typeof response === "object" &&
+			Array.isArray(response?.choices) &&
+			typeof response?.created === "number" &&
+			typeof response?.id === "string" &&
+			typeof response?.model === "string" &&
+			typeof response?.usage === "object"
+		) {
+			return response;
+		}
+
+		throw new InferenceOutputError("Expected ChatCompletionOutput");
+	}
+}
+
+export class CentMLTextGenerationTask extends BaseTextGenerationTask {
+	constructor() {
+		super("centml", CENTML_API_BASE_URL);
+	}
+
+	override makeRoute(): string {
+		return "openai/v1/completions";
+	}
+
+	override preparePayload(params: BodyParams): Record<string, unknown> {
+		const { args, model } = params;
+		return {
+			...args,
+			model,
+			api_key: args.accessToken,
+		};
+	}
+
+	override async getResponse(response: TextGenerationOutput): Promise<TextGenerationOutput> {
+		if (typeof response === "object" && typeof response?.generated_text === "string") {
+			return response;
+		}
+
+		throw new InferenceOutputError("Expected TextGenerationOutput");
+	}
+}

packages/inference/src/providers/consts.ts

@@ -21,6 +21,14 @@ export const HARDCODED_MODEL_INFERENCE_MAPPING: Record<
 	 */
 	"black-forest-labs": {},
 	cerebras: {},
+	centml: {
+		"meta-llama/Llama-3.2-3B-Instruct": {
+			hfModelId: "meta-llama/Llama-3.2-3B-Instruct",
+			providerId: "meta-llama/Llama-3.2-3B-Instruct",  // CentML expects same id
+			status: "live",          // or "staging" if you prefer the warning
+			task: "conversational"   // <-- WidgetType from @huggingface/tasks
+		}
+	},
 	cohere: {},
 	"fal-ai": {},
 	"featherless-ai": {},

packages/inference/src/types.ts

@@ -40,6 +40,7 @@ export type InferenceTask = Exclude<PipelineType, "other"> | "conversational";
 export const INFERENCE_PROVIDERS = [
 	"black-forest-labs",
 	"cerebras",
+	"centml",
 	"cohere",
 	"fal-ai",
 	"featherless-ai",