feat: Model Discovery

packages/core/src/context.ts

@@ -1,6 +1,6 @@
 import { createLogger } from '@sap-cloud-sdk/util';
 import {
-  Destination,
+  HttpDestination,
   Service,
   ServiceCredentials,
   getServiceBinding,
@@ -18,7 +18,7 @@ let aiCoreServiceBinding: Service | undefined;
  * Returns a destination object from AI Core service binding.
  * @returns The destination object.
  */
-export async function getAiCoreDestination(): Promise<Destination> {
+export async function getAiCoreDestination(): Promise<HttpDestination> {
   if (!aiCoreServiceBinding) {
     aiCoreServiceBinding =
       getAiCoreServiceKeyFromEnv() || getServiceBinding('aicore');
@@ -29,12 +29,12 @@ export async function getAiCoreDestination(): Promise<Destination> {
     }
   }
 
-  const aiCoreDestination = await transformServiceBindingToDestination(
+  const aiCoreDestination = (await transformServiceBindingToDestination(
     aiCoreServiceBinding,
     {
       useCache: true
     }
-  );
+  )) as HttpDestination;
   return aiCoreDestination;
 }
 

packages/core/src/http-client.ts

@@ -54,19 +54,17 @@ export interface EndpointOptions {
  * @param requestConfig - The request configuration.
  * @returns The {@link HttpResponse} from the AI Core service.
  */
-export async function executeRequest<Data extends BaseLlmParameters>(
+export async function executeRequest(
   endpointOptions: EndpointOptions,
-  data: Data,
+  data: any,
   requestConfig?: CustomRequestConfig
 ): Promise<HttpResponse> {
   const aiCoreDestination = await getAiCoreDestination();
-  // eslint-disable-next-line @typescript-eslint/no-unused-vars
-  const { deploymentConfiguration, ...body } = data;
   const { url, apiVersion } = endpointOptions;
 
   const mergedRequestConfig = {
     ...mergeWithDefaultRequestConfig(apiVersion, requestConfig),
-    data: JSON.stringify(body)
+    data: JSON.stringify(data)
   };
 
   const targetUrl = aiCoreDestination.url + `/v2/${removeLeadingSlashes(url)}`;

packages/core/src/openapi-request-builder.ts

@@ -27,13 +27,9 @@ export class OpenApiRequestBuilder<
   async executeRaw(): Promise<HttpResponse> {
     const { url, data, ...rest } = await this.requestConfig();
     // TODO: Remove explicit url! once we updated the type in the Cloud SDK, since url is always defined.
-    return executeRequest(
-      { url: url! },
-      { deploymentConfiguration: {}, ...data },
-      {
-        ...rest
-      }
-    );
+    return executeRequest({ url: url! }, data, {
+      ...rest
+    });
   }
 
   /**

packages/gen-ai-hub/package.json

@@ -28,6 +28,7 @@
   },
   "dependencies": {
     "@sap-ai-sdk/core": "workspace:^",
+    "@sap-ai-sdk/ai-core": "workspace:^",
     "@sap-cloud-sdk/http-client": "^3.18.1",
     "@sap-cloud-sdk/connectivity": "^3.18.1",
     "@sap-cloud-sdk/util": "^3.18.1",

packages/gen-ai-hub/src/client/openai/openai-client.test.ts

@@ -1,29 +1,24 @@
 import nock from 'nock';
-import { BaseLlmParametersWithDeploymentId } from '@sap-ai-sdk/core';
 import {
   mockClientCredentialsGrantCall,
   mockInference,
   parseMockResponse
 } from '../../../../../test-util/mock-http.js';
 import {
   OpenAiChatCompletionOutput,
-  OpenAiChatCompletionParameters,
   OpenAiChatMessage,
   OpenAiEmbeddingOutput,
   OpenAiEmbeddingParameters
 } from './openai-types.js';
 import { OpenAiClient } from './openai-client.js';
 
 describe('openai client', () => {
-  const deploymentConfiguration: BaseLlmParametersWithDeploymentId = {
-    deploymentId: 'deployment-id'
-  };
   const chatCompletionEndpoint = {
-    url: `inference/deployments/${deploymentConfiguration.deploymentId}/chat/completions`,
+    url: 'inference/deployments/1234/chat/completions',
     apiVersion: '2024-02-01'
   };
   const embeddingsEndpoint = {
-    url: `inference/deployments/${deploymentConfiguration.deploymentId}/embeddings`,
+    url: 'inference/deployments/1234/embeddings',
     apiVersion: '2024-02-01'
   };
 
@@ -47,18 +42,15 @@ describe('openai client', () => {
           }
         ] as OpenAiChatMessage[]
       };
-      const request: OpenAiChatCompletionParameters = {
-        ...prompt,
-        deploymentConfiguration
-      };
+
       const mockResponse = parseMockResponse<OpenAiChatCompletionOutput>(
         'openai',
         'openai-chat-completion-success-response.json'
       );
 
       mockInference(
         {
-          data: request
+          data: prompt
         },
         {
           data: mockResponse,
@@ -67,24 +59,24 @@ describe('openai client', () => {
         chatCompletionEndpoint
       );
 
-      const response = await client.chatCompletion(request);
+      const response = await client.chatCompletion(
+        'gpt-35-turbo',
+        prompt,
+        '1234'
+      );
       expect(response).toEqual(mockResponse);
     });
 
     it('throws on bad request', async () => {
       const prompt = { messages: [] };
-      const request: OpenAiChatCompletionParameters = {
-        ...prompt,
-        deploymentConfiguration
-      };
       const mockResponse = parseMockResponse(
         'openai',
         'openai-error-response.json'
       );
 
       mockInference(
         {
-          data: request
+          data: prompt
         },
         {
           data: mockResponse,
@@ -93,50 +85,50 @@ describe('openai client', () => {
         chatCompletionEndpoint
       );
 
-      expect(client.chatCompletion(request)).rejects.toThrow();
+      await expect(
+        client.chatCompletion('gpt-4', prompt, '1234')
+      ).rejects.toThrow('status code 400');
     });
   });
 
   describe('embeddings', () => {
     it('parses a successful response', async () => {
-      const prompt = { input: ['AI is fascinating'] };
-      const request: OpenAiEmbeddingParameters = {
-        ...prompt,
-        deploymentConfiguration
-      };
+      const prompt = {
+        input: ['AI is fascinating']
+      } as OpenAiEmbeddingParameters;
       const mockResponse = parseMockResponse<OpenAiEmbeddingOutput>(
         'openai',
         'openai-embeddings-success-response.json'
       );
 
       mockInference(
         {
-          data: request
+          data: prompt
         },
         {
           data: mockResponse,
           status: 200
         },
         embeddingsEndpoint
       );
-      const response = await client.embeddings(request);
+      const response = await client.embeddings(
+        'text-embedding-ada-002',
+        prompt,
+        '1234'
+      );
       expect(response).toEqual(mockResponse);
     });
 
     it('throws on bad request', async () => {
       const prompt = { input: [] };
-      const request: OpenAiEmbeddingParameters = {
-        ...prompt,
-        deploymentConfiguration
-      };
       const mockResponse = parseMockResponse(
         'openai',
         'openai-error-response.json'
       );
 
       mockInference(
         {
-          data: request
+          data: prompt
         },
         {
           data: mockResponse,
@@ -145,7 +137,9 @@ describe('openai client', () => {
         embeddingsEndpoint
       );
 
-      expect(client.embeddings(request)).rejects.toThrow();
+      await expect(
+        client.embeddings('text-embedding-3-large', prompt, '1234')
+      ).rejects.toThrow('status code 400');
     });
   });
 });

packages/gen-ai-hub/src/client/openai/openai-client.ts

@@ -1,60 +1,107 @@
+import { HttpRequestConfig } from '@sap-cloud-sdk/http-client';
+import { CustomRequestConfig, executeRequest } from '@sap-ai-sdk/core';
 import {
-  BaseLlmParameters,
-  CustomRequestConfig,
-  executeRequest
-} from '@sap-ai-sdk/core';
-import { BaseClient } from '../interface.js';
+  DeploymentResolver,
+  resolveDeployment
+} from '../../utils/deployment-resolver.js';
 import {
   OpenAiChatCompletionParameters,
   OpenAiEmbeddingParameters,
   OpenAiEmbeddingOutput,
-  OpenAiChatCompletionOutput
+  OpenAiChatCompletionOutput,
+  OpenAiChatModel,
+  OpenAiEmbeddingModel
 } from './openai-types.js';
 
 const apiVersion = '2024-02-01';
 
 /**
- * OpenAI GPT Client.
+ * OpenAI Client.
  */
-export class OpenAiClient implements BaseClient<BaseLlmParameters> {
+export class OpenAiClient {
   /**
    * Creates a completion for the chat messages.
+   * @param model - The model to use for the chat completion.
    * @param data - The input parameters for the chat completion.
+   * @param deploymentResolver - A deployment id or a function to retrieve it.
    * @param requestConfig - The request configuration.
    * @returns The completion result.
    */
   async chatCompletion(
+    model: OpenAiChatModel | { name: OpenAiChatModel; version: string },
     data: OpenAiChatCompletionParameters,
+    deploymentResolver?: DeploymentResolver,
     requestConfig?: CustomRequestConfig
   ): Promise<OpenAiChatCompletionOutput> {
+    const deploymentId = await resolveOpenAiDeployment(
+      model,
+      deploymentResolver
+    );
     const response = await executeRequest(
       {
-        url: `/inference/deployments/${data.deploymentConfiguration.deploymentId}/chat/completions`,
+        url: `/inference/deployments/${deploymentId}/chat/completions`,
         apiVersion
       },
       data,
-      requestConfig
+      mergeRequestConfig(requestConfig)
     );
     return response.data;
   }
   /**
    * Creates an embedding vector representing the given text.
-   * @param data - The input parameters for the chat completion.
+   * @param model - The model to use for the embedding computation.
+   * @param data - The text to embed.
+   * @param deploymentResolver - A deployment id or a function to retrieve it.
    * @param requestConfig - The request configuration.
    * @returns The completion result.
    */
   async embeddings(
+    model:
+      | OpenAiEmbeddingModel
+      | { name: OpenAiEmbeddingModel; version: string },
     data: OpenAiEmbeddingParameters,
+    deploymentResolver?: DeploymentResolver,
     requestConfig?: CustomRequestConfig
   ): Promise<OpenAiEmbeddingOutput> {
+    const deploymentId = await resolveOpenAiDeployment(
+      model,
+      deploymentResolver
+    );
     const response = await executeRequest(
-      {
-        url: `/inference/deployments/${data.deploymentConfiguration.deploymentId}/embeddings`,
-        apiVersion
-      },
+      { url: `/inference/deployments/${deploymentId}/embeddings`, apiVersion },
       data,
-      requestConfig
+      mergeRequestConfig(requestConfig)
     );
     return response.data;
   }
 }
+
+async function resolveOpenAiDeployment(
+  model: string | { name: string; version: string },
+  resolver?: DeploymentResolver
+) {
+  if (typeof resolver === 'string') {
+    return resolver;
+  }
+  const llm =
+    typeof model === 'string' ? { name: model, version: 'latest' } : model;
+  const deployment = await resolveDeployment({
+    scenarioId: 'foundation-models',
+    executableId: 'azure-openai',
+    model: llm
+  });
+  return deployment.id;
+}
+
+function mergeRequestConfig(
+  requestConfig?: CustomRequestConfig
+): HttpRequestConfig {
+  return {
+    method: 'POST',
+    headers: {
+      'content-type': 'application/json'
+    },
+    params: { 'api-version': apiVersion },
+    ...requestConfig
+  };
+}