Merge pull request #6023 from continuedev/pe/watsonx-vi

feat: use correct deployment for azure

Author: sestinj

.github/workflows/pr_checks.yaml

@@ -435,6 +435,8 @@ jobs:
           MISTRAL_API_KEY: ${{ secrets.MISTRAL_API_KEY }}
           AZURE_OPENAI_API_KEY: ${{ secrets.AZURE_OPENAI_API_KEY }}
           AZURE_FOUNDRY_API_KEY: ${{ secrets.AZURE_FOUNDRY_API_KEY }}
+          AZURE_FOUNDRY_MISTRAL_SMALL_API_KEY: ${{ secrets.AZURE_FOUNDRY_MISTRAL_SMALL_API_KEY }}
+          AZURE_OPENAI_GPT41_API_KEY: ${{ secrets.AZURE_OPENAI_GPT41_API_KEY }}
 
   package-tests:
     runs-on: ubuntu-latest

packages/openai-adapters/src/apis/Azure.ts

@@ -0,0 +1,108 @@
+import { OpenAI } from "openai/index";
+import {
+  ChatCompletionChunk,
+  ChatCompletionCreateParams,
+  ChatCompletionCreateParamsStreaming,
+} from "openai/resources/index";
+import { z } from "zod";
+import { AzureConfigSchema } from "../types.js";
+import { customFetch } from "../util.js";
+import { OpenAIApi } from "./OpenAI.js";
+
+export class AzureApi extends OpenAIApi {
+  constructor(private azureConfig: z.infer<typeof AzureConfigSchema>) {
+    super({
+      ...azureConfig,
+      provider: "openai",
+    });
+
+    this.openai = new OpenAI({
+      apiKey: azureConfig.apiKey,
+      baseURL: this._getAzureBaseURL(azureConfig),
+      defaultQuery: azureConfig.env?.apiVersion
+        ? { "api-version": azureConfig.env.apiVersion }
+        : {},
+      fetch: customFetch(azureConfig.requestOptions),
+    });
+  }
+
+  /**
+   * Default is `azure-openai`, but previously was `azure`
+   * @param apiType
+   * @returns
+   */
+  private _isAzureOpenAI(apiType?: string): boolean {
+    return apiType === "azure-openai" || apiType === "azure";
+  }
+
+  private _getAzureBaseURL(config: z.infer<typeof AzureConfigSchema>): string {
+    const baseURL = new URL(this.apiBase).toString().replace(/\/$/, "");
+
+    // Default is `azure-openai` in docs, but previously was `azure`
+    if (this._isAzureOpenAI(config.env?.apiType)) {
+      if (!config.env?.deployment) {
+        throw new Error(
+          "Azure deployment is required if `apiType` is `azure-openai` or `azure`",
+        );
+      }
+
+      return `${baseURL}/openai/deployments/${config.env.deployment}`;
+    }
+
+    return baseURL;
+  }
+
+  /**
+   * Filters out empty text content parts from messages.
+   *
+   * Azure models may not support empty content parts, which can cause issues.
+   * This function removes any text content parts that are empty or contain only whitespace.
+   */
+  private _filterEmptyContentParts<T extends ChatCompletionCreateParams>(
+    body: T,
+  ): T {
+    const result = { ...body };
+
+    result.messages = result.messages.map((message: any) => {
+      if (Array.isArray(message.content)) {
+        const filteredContent = message.content.filter((part: any) => {
+          return !(
+            part.type === "text" &&
+            (!part.text || part.text.trim() === "")
+          );
+        });
+        return {
+          ...message,
+          content:
+            filteredContent.length > 0 ? filteredContent : message.content,
+        };
+      }
+      return message;
+    }) as any;
+
+    return result;
+  }
+
+  modifyChatBody<T extends ChatCompletionCreateParams>(body: T): T {
+    let modifiedBody = super.modifyChatBody(body);
+    modifiedBody = this._filterEmptyContentParts(modifiedBody);
+    return modifiedBody;
+  }
+
+  async *chatCompletionStream(
+    body: ChatCompletionCreateParamsStreaming,
+    signal: AbortSignal,
+  ): AsyncGenerator<ChatCompletionChunk, any, unknown> {
+    const response = await this.openai.chat.completions.create(
+      this.modifyChatBody(body),
+      { signal },
+    );
+
+    for await (const result of response) {
+      // Skip chunks with no choices (common with Azure content filtering)
+      if (result.choices && result.choices.length > 0) {
+        yield result;
+      }
+    }
+  }
+}

packages/openai-adapters/src/apis/OpenAI.ts

@@ -12,7 +12,7 @@ import {
   Model,
 } from "openai/resources/index";
 import { z } from "zod";
-import { AzureConfigSchema, OpenAIConfigSchema } from "../types.js";
+import { OpenAIConfigSchema } from "../types.js";
 import { customFetch } from "../util.js";
 import {
   BaseLlmApi,
@@ -25,19 +25,14 @@ export class OpenAIApi implements BaseLlmApi {
   openai: OpenAI;
   apiBase: string = "https://api.openai.com/v1/";
 
-  constructor(
-    protected config: z.infer<
-      typeof OpenAIConfigSchema | typeof AzureConfigSchema
-    >,
-  ) {
+  constructor(protected config: z.infer<typeof OpenAIConfigSchema>) {
     this.apiBase = config.apiBase ?? this.apiBase;
     this.openai = new OpenAI({
       apiKey: config.apiKey,
       baseURL: this.apiBase,
       fetch: customFetch(config.requestOptions),
     });
   }
-
   modifyChatBody<T extends ChatCompletionCreateParams>(body: T): T {
     // o-series models - only apply for official OpenAI API
     const isOfficialOpenAIAPI = this.apiBase === "https://api.openai.com/v1/";
@@ -74,6 +69,7 @@ export class OpenAIApi implements BaseLlmApi {
     );
     return response;
   }
+
   async *chatCompletionStream(
     body: ChatCompletionCreateParamsStreaming,
     signal: AbortSignal,

packages/openai-adapters/src/index.ts

@@ -1,6 +1,7 @@
 import dotenv from "dotenv";
 import { z } from "zod";
 import { AnthropicApi } from "./apis/Anthropic.js";
+import { AzureApi } from "./apis/Azure.js";
 import { CohereApi } from "./apis/Cohere.js";
 import { DeepSeekApi } from "./apis/DeepSeek.js";
 import { GeminiApi } from "./apis/Gemini.js";
@@ -29,8 +30,9 @@ function openAICompatible(
 export function constructLlmApi(config: LLMConfig): BaseLlmApi | undefined {
   switch (config.provider) {
     case "openai":
-    case "azure":
       return new OpenAIApi(config);
+    case "azure":
+      return new AzureApi(config);
     case "cohere":
       return new CohereApi(config);
     case "anthropic":

packages/openai-adapters/src/test/main.test.ts

@@ -4,11 +4,16 @@ import { DEEPSEEK_API_BASE } from "../apis/DeepSeek.js";
 import { INCEPTION_API_BASE } from "../apis/Inception.js";
 import { OpenAIApi } from "../apis/OpenAI.js";
 import { constructLlmApi } from "../index.js";
-import { getLlmApi, testChat, testCompletion, testEmbed } from "./util.js";
+import { getLlmApi, testChat, testEmbed } from "./util.js";
 
 dotenv.config();
 
-function testConfig(config: ModelConfig) {
+export interface TestConfigOptions {
+  skipTools: boolean;
+}
+
+function testConfig(_config: ModelConfig & { options?: TestConfigOptions }) {
+  const { options, ...config } = _config;
   const model = config.model;
   const api = getLlmApi({
     provider: config.provider as any,
@@ -17,16 +22,12 @@ function testConfig(config: ModelConfig) {
     env: config.env,
   });
 
-  if (false) {
-    testCompletion(api, model);
-  }
-
   if (
     ["chat", "summarize", "edit", "apply"].some((role) =>
       config.roles?.includes(role as any),
     )
   ) {
-    testChat(api, model);
+    testChat(api, model, options);
   }
 
   if (config.roles?.includes("embed")) {
@@ -38,7 +39,7 @@ function testConfig(config: ModelConfig) {
   }
 }
 
-const TESTS: Omit<ModelConfig, "name">[] = [
+const TESTS: Omit<ModelConfig & { options?: TestConfigOptions }, "name">[] = [
   {
     provider: "openai",
     model: "gpt-4o",
@@ -106,6 +107,26 @@ const TESTS: Omit<ModelConfig, "name">[] = [
   //   apiKey: process.env.COHERE_API_KEY!,
   //   roles: ["rerank"],
   // },
+  {
+    provider: "azure",
+    model: "gpt-4.1",
+    apiBase: "https://continue-openai.openai.azure.com",
+    apiKey: process.env.AZURE_OPENAI_GPT41_API_KEY,
+    env: {
+      deployment: "gpt-4.1",
+      apiVersion: "2024-02-15-preview",
+      apiType: "azure-openai",
+    },
+    roles: ["chat"],
+  },
+  {
+    provider: "azure",
+    model: "mistral-small-2503",
+    apiBase: "https://nate-0276-resource.services.ai.azure.com/models",
+    apiKey: process.env.AZURE_FOUNDRY_MISTRAL_SMALL_API_KEY,
+    roles: ["chat"],
+    options: { skipTools: true },
+  },
 ];
 
 TESTS.forEach((config) => {