From d7dca5eb9ebc00f8470af91d392a21fb6d89ecf3 Mon Sep 17 00:00:00 2001
From: Nils Barring <nils.barring@nscale.com>
Date: Mon, 4 Aug 2025 16:16:15 +0800
Subject: [PATCH 1/2] add v1/completions support to nscale inference provider

---
 .../inference/src/lib/getProviderHelper.ts    |  1 +
 packages/inference/src/providers/nscale.ts    | 48 ++++++++++++++++++-
 .../inference/test/InferenceClient.spec.ts    | 28 +++++++++--
 3 files changed, 73 insertions(+), 4 deletions(-)

diff --git a/packages/inference/src/lib/getProviderHelper.ts b/packages/inference/src/lib/getProviderHelper.ts
index 72cc35bf62..bf439aca6d 100644
--- a/packages/inference/src/lib/getProviderHelper.ts
+++ b/packages/inference/src/lib/getProviderHelper.ts
@@ -130,6 +130,7 @@ export const PROVIDERS: Record<InferenceProvider, Partial<Record<InferenceTask,
 	nscale: {
 		"text-to-image": new Nscale.NscaleTextToImageTask(),
 		conversational: new Nscale.NscaleConversationalTask(),
+		"text-generation": new Nscale.NscaleTextGenerationTask(),
 	},
 	openai: {
 		conversational: new OpenAI.OpenAIConversationalTask(),
diff --git a/packages/inference/src/providers/nscale.ts b/packages/inference/src/providers/nscale.ts
index a4920973a6..ecb6e78490 100644
--- a/packages/inference/src/providers/nscale.ts
+++ b/packages/inference/src/providers/nscale.ts
@@ -15,13 +15,26 @@
  * Thanks!
  */
 import type { TextToImageInput } from "@huggingface/tasks";
+import type { TextGenerationOutput, TextGenerationOutputFinishReason } from "@huggingface/tasks";
 import type { BodyParams } from "../types.js";
 import { omit } from "../utils/omit.js";
-import { BaseConversationalTask, TaskProviderHelper, type TextToImageTaskHelper } from "./providerHelper.js";
+import {
+	BaseConversationalTask,
+	BaseTextGenerationTask,
+	TaskProviderHelper,
+	type TextToImageTaskHelper,
+} from "./providerHelper.js";
 import { InferenceClientProviderOutputError } from "../errors.js";
 
 const NSCALE_API_BASE_URL = "https://inference.api.nscale.com";
 
+interface NscaleTextGenerationOutput extends Omit<TextGenerationOutput, "choices"> {
+	choices: Array<{
+		text: string;
+		finish_reason: TextGenerationOutputFinishReason;
+	}>;
+}
+
 interface NscaleCloudBase64ImageGeneration {
 	data: Array<{
 		b64_json: string;
@@ -34,6 +47,39 @@ export class NscaleConversationalTask extends BaseConversationalTask {
 	}
 }
 
+export class NscaleTextGenerationTask extends BaseTextGenerationTask {
+	constructor() {
+		super("nscale", NSCALE_API_BASE_URL);
+	}
+
+	override makeRoute(): string {
+		return "v1/completions";
+	}
+
+	override preparePayload(params: BodyParams): Record<string, unknown> {
+		return {
+			model: params.model,
+			prompt: params.args.inputs,
+			...(params.args.parameters || {}),
+		};
+	}
+
+	override async getResponse(response: NscaleTextGenerationOutput): Promise<TextGenerationOutput> {
+		if (
+			typeof response === "object" &&
+			"choices" in response &&
+			Array.isArray(response?.choices) &&
+			response.choices.length > 0 &&
+			typeof response.choices[0]?.text === "string"
+		) {
+			return {
+				generated_text: response.choices[0].text,
+			};
+		}
+		throw new InferenceClientProviderOutputError("Received malformed response from Nscale text generation API");
+	}
+}
+
 export class NscaleTextToImageTask extends TaskProviderHelper implements TextToImageTaskHelper {
 	constructor() {
 		super("nscale", NSCALE_API_BASE_URL);
diff --git a/packages/inference/test/InferenceClient.spec.ts b/packages/inference/test/InferenceClient.spec.ts
index d5cefcc60e..1f1a8c287c 100644
--- a/packages/inference/test/InferenceClient.spec.ts
+++ b/packages/inference/test/InferenceClient.spec.ts
@@ -22,7 +22,7 @@ if (!env.HF_TOKEN) {
 	console.warn("Set HF_TOKEN in the env to run the tests for better rate limits");
 }
 
-describe.skip("InferenceClient", () => {
+describe("InferenceClient", () => {
 	// Individual tests can be ran without providing an api key, however running all tests without an api key will result in rate limiting error.
 
 	describe("backward compatibility", () => {
@@ -1918,14 +1918,21 @@ describe.skip("InferenceClient", () => {
 				"meta-llama/Llama-3.1-8B-Instruct": {
 					provider: "nscale",
 					hfModelId: "meta-llama/Llama-3.1-8B-Instruct",
-					providerId: "nscale",
+					providerId: "meta-llama/Llama-3.1-8B-Instruct",
 					status: "live",
 					task: "conversational",
 				},
+				"mistralai/Devstral-Small-2505": {
+					provider: "nscale",
+					hfModelId: "mistralai/Devstral-Small-2505",
+					providerId: "mistralai/Devstral-Small-2505",
+					status: "staging",
+					task: "text-generation",
+				},
 				"black-forest-labs/FLUX.1-schnell": {
 					provider: "nscale",
 					hfModelId: "black-forest-labs/FLUX.1-schnell",
-					providerId: "flux-schnell",
+					providerId: "black-forest-labs/FLUX.1-schnell",
 					status: "live",
 					task: "text-to-image",
 				},
@@ -1969,6 +1976,21 @@ describe.skip("InferenceClient", () => {
 				});
 				expect(res).toBeInstanceOf(Blob);
 			});
+
+			it("textGeneration", async () => {
+				const res = await client.textGeneration({
+					model: "mistralai/Devstral-Small-2505",
+					provider: "nscale",
+					inputs: "1+1=",
+					parameters: {
+						temperature: 0,
+						max_tokens: 1,
+					},
+				});
+
+				expect(res.generated_text.length > 0);
+				expect(res.generated_text).toContain("2");
+			});
 		},
 		TIMEOUT
 	);

From f6a8ba2951e732551495d0a58950b76ea92d5790 Mon Sep 17 00:00:00 2001
From: Nils Barring <nils.barring@nscale.com>
Date: Mon, 4 Aug 2025 17:10:31 +0800
Subject: [PATCH 2/2] added skip tests back

---
 packages/inference/test/InferenceClient.spec.ts | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/packages/inference/test/InferenceClient.spec.ts b/packages/inference/test/InferenceClient.spec.ts
index 1f1a8c287c..407bfe176e 100644
--- a/packages/inference/test/InferenceClient.spec.ts
+++ b/packages/inference/test/InferenceClient.spec.ts
@@ -22,7 +22,7 @@ if (!env.HF_TOKEN) {
 	console.warn("Set HF_TOKEN in the env to run the tests for better rate limits");
 }
 
-describe("InferenceClient", () => {
+describe.skip("InferenceClient", () => {
 	// Individual tests can be ran without providing an api key, however running all tests without an api key will result in rate limiting error.
 
 	describe("backward compatibility", () => {