fix chatcompletion payload

Author: Fabien Ric

packages/inference/src/providers/ovhcloud.ts

@@ -24,23 +24,10 @@ import type {
 import { InferenceOutputError } from "../lib/InferenceOutputError";
 import type { BodyParams } from "../types";
 import { omit } from "../utils/omit";
+import type { TextGenerationInput } from "@huggingface/tasks";
 
 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;
@@ -54,21 +41,25 @@ 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 preparePayload(params: BodyParams<TextGenerationInput>): 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,
+		};
 	}
 
 	override async getResponse(response: OvhCloudTextCompletionOutput): Promise<TextGenerationOutput> {

packages/inference/test/InferenceClient.spec.ts

@@ -1710,12 +1710,10 @@ describe.concurrent("InferenceClient", () => {
 					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,
-					},
+					seed: 42,
+					temperature: 0,
+					top_p: 0.01,
+					max_tokens: 1,
 				});
 				expect(res.choices && res.choices.length > 0);
 				const completion = res.choices[0].message?.content;
@@ -1728,12 +1726,10 @@ describe.concurrent("InferenceClient", () => {
 					provider: "ovhcloud",
 					messages: [{ role: "user", content: "A, B, C, " }],
 					stream: true,
-					parameters: {
-						seed: 42,
-						temperature: 0,
-						top_p: 0.01,
-						max_new_tokens: 1,
-					},
+					seed: 42,
+					temperature: 0,
+					top_p: 0.01,
+					max_tokens: 1,
 				}) as AsyncGenerator<ChatCompletionStreamOutput>;
 
 				let fullResponse = "";