SITES-26591 [Import Assistant] Expand GPT client to user chat endpoint

- PR suggestion (more in options, better README)
adobe · Nov 5, 2024 · 4091cb8 · 4091cb8
1 parent f231622
commit 4091cb8
Show file tree

Hide file tree

Showing 3 changed files with 85 additions and 67 deletions.
diff --git a/packages/spacecat-shared-gpt-client/README.md b/packages/spacecat-shared-gpt-client/README.md
@@ -45,6 +45,9 @@ try {
 #### Via Capability Execution endpoint
 
 ```javascript
+/**
+ *  Fetch insights using the Firefall's capability execution endpoint.
+ */
 async function fetchInsights(prompt) {
   try {
     const client = FirefallClient.createFrom({
@@ -60,7 +63,6 @@ async function fetchInsights(prompt) {
       log: console,
     });
 
-    // Internally, use Firefall's .../capability_execution/job endpoint.
     const insights = await client.fetchCapabilityExecution(prompt);
     console.log('Insights:', insights);
   } catch (error) {
@@ -74,13 +76,15 @@ fetchInsights('How can we improve customer satisfaction?');
 #### Via Chat Completions endpoint
 
 ```javascript
-async function fetchInsights(prompt) {
+/**
+ *  Fetch completions using the Firefall's chat completions endpoint.
+ */
+async function fetchCompletions(prompt) {
   try {
     const client = FirefallClient.createFrom({
       env: {
         FIREFALL_API_ENDPOINT: 'https://api.firefall.example.com',
         FIREFALL_API_KEY: 'yourApiKey',
-        FIREFALL_API_CAPABILITY_NAME: 'yourCapabilityName',
         IMS_HOST: 'ims.example.com',
         IMS_CLIENT_ID: 'yourClientId',
         IMS_CLIENT_CODE: 'yourClientCode',
@@ -89,18 +93,19 @@ async function fetchInsights(prompt) {
       log: console,
     });
     const options = {
-      imageUrls: ['data:image/png;base64,iVBORw0KGgoAAAA...=']
+      imageUrls: ['data:image/png;base64,iVBORw0KGgoAAAA...='],
+      model:'gpt-4-vision',
+      responseFormat: undefined,
     };
 
-    // // Internally, use Firefall's .../chat/completions endpoint.
-    const insights = await client.fetchChatCompletion(prompt, { options });
-    console.log('Insights:', insights);
+    const response = await client.fetchChatCompletion(prompt, { options });
+    console.log('Response:', JSON.stringify(response));
   } catch (error) {
-    console.error('Failed to fetch insights:', error.message);
+    console.error('Failed to fetch chat completion:', error.message);
   }
 }
 
-fetchInsights('Identify all food items in this image', { imageUrls: ['data:image/png;base64,iVBORw0KGgoAAAA...='] });
+fetchCompletions('Identify all food items in this image', { imageUrls: ['data:image/png;base64,iVBORw0KGgoAAAA...='] });
 ```
 
 Ensure that you replace `'path/to/firefall-client'` with the actual path to the `FirefallClient` class in your project and adjust the configuration parameters according to your Firefall API credentials.

diff --git a/packages/spacecat-shared-gpt-client/src/clients/firefall-client.js b/packages/spacecat-shared-gpt-client/src/clients/firefall-client.js
@@ -33,6 +33,10 @@ function validateChatCompletionResponse(response) {
 }
 
 export default class FirefallClient {
+  static STAGE_FIREFALL_API_ENDPOINT = 'http://firefall-stage.adobe.io';
+
+  static PROD_FIREFALL_API_ENDPOINT = 'http://firefall.adobe.io/';
+
   static createFrom(context) {
     const { log = console } = context;
     const imsClient = ImsClient.createFrom(context);
@@ -96,7 +100,7 @@ export default class FirefallClient {
     this.log.debug(`${message}: took ${duration}ms`);
   }
 
-  async #submitPrompt(prompt, body, path) {
+  async #submitPrompt(body, path) {
     const apiAuth = await this.#getApiAuth();
 
     const url = createUrl(`${this.config.apiEndpoint}${path}`);
@@ -168,16 +172,13 @@ export default class FirefallClient {
    * @param prompt The text prompt to provide to Firefall
    * @param options The options for the call:
    *                - imageUrls: An array of URLs of the images to provide to Firefall
-   * @returns {Promise<*>}
+   * @returns {Object} - AI response
    */
   async fetchChatCompletion(prompt, options = {}) {
-    if (!hasText(prompt)) {
-      throw new Error('Invalid prompt received');
-    }
-    const hasImageUrls = options?.imageUrls && options.imageUrls.length > 0;
+    const getBody = () => {
+      const { imageUrls, responseFormat, model: llmModel = 'gpt-4-turbo' } = options || {};
+      const hasImageUrls = imageUrls && imageUrls.length > 0;
 
-    try {
-      const startTime = process.hrtime.bigint();
       const userRole = {
         role: 'user',
         content: [
@@ -187,27 +188,11 @@ export default class FirefallClient {
           },
         ],
       };
-      const body = {
-        llm_metadata: {
-          model_name: this.config.FIREFALL_API_CAPABILITY_NAME,
-          llm_type: 'azure_chat_openai',
-        },
-        messages: [
-          {
-            role: 'system',
-            content: 'You are a helpful assistant designed to output JSON.',
-          },
-          userRole,
-        ],
-      };
+
       if (hasImageUrls) {
-        if (!Array.isArray(options?.imageUrls)) {
+        if (!Array.isArray(imageUrls)) {
           throw new Error('imageUrls must be an array.');
         }
-        // Warn if model might not handle images.
-        if (!this.config.capabilityName.includes('vision') && this.log?.warn) {
-          this.log.warn(`Image URLs were provided but capability (${this.config.capabilityName}) may not handle vision prompts. Continuing...`);
-        }
         options.imageUrls.forEach((imageUrl) => {
           userRole.content.push({
             type: 'image_url',
@@ -217,34 +202,65 @@ export default class FirefallClient {
           });
         });
       }
-      const chatSubmissionResponse = await this.#submitPrompt(
-        prompt,
+
+      const body = {
+        llm_metadata: {
+          model_name: llmModel,
+          llm_type: 'azure_chat_openai',
+        },
+        messages: [
+          userRole,
+        ],
+      };
+      if (responseFormat === 'json_object') {
+        body.response_format = {
+          type: 'json_object',
+        };
+        body.messages.push({
+          role: 'system',
+          content: 'You are a helpful assistant designed to output JSON.',
+        });
+      }
+
+      return body;
+    };
+
+    if (!hasText(prompt)) {
+      throw new Error('Invalid prompt received');
+    }
+
+    let chatSubmissionResponse;
+    try {
+      const startTime = process.hrtime.bigint();
+      const body = getBody();
+
+      chatSubmissionResponse = await this.#submitPrompt(
         JSON.stringify(body),
         '/v2/chat/completions',
       );
       this.#logDuration('Firefall API Chat Completion call', startTime);
-
-      if (!validateChatCompletionResponse(chatSubmissionResponse)) {
-        this.log.error(
-          'Could not obtain data from Firefall: Invalid response format.',
-        );
-        throw new Error('Invalid response format.');
-      }
-      if (!chatSubmissionResponse.choices.some((ch) => hasText(ch?.message?.content))) {
-        throw new Error('Prompt completed but no output was found.');
-      }
-
-      return chatSubmissionResponse;
     } catch (error) {
       this.log.error('Error while fetching data from Firefall chat API: ', error.message);
       throw error;
     }
+
+    if (!validateChatCompletionResponse(chatSubmissionResponse)) {
+      this.log.error(
+        'Could not obtain data from Firefall: Invalid response format.',
+      );
+      throw new Error('Invalid response format.');
+    }
+    if (!chatSubmissionResponse.choices.some((ch) => hasText(ch?.message?.content))) {
+      throw new Error('Prompt completed but no output was found.');
+    }
+
+    return chatSubmissionResponse;
   }
 
   /**
    * Fetches data from Firefall API.
    * @param prompt The text prompt to provide to Firefall
-   * @returns {Promise<*>}
+   * @returns {string} - AI response
    */
   async fetchCapabilityExecution(prompt) {
     if (!hasText(prompt)) {
@@ -260,7 +276,7 @@ export default class FirefallClient {
       });
       const path = '/v2/capability_execution/job';
 
-      const jobSubmissionResponse = await this.#submitPrompt(prompt, body, path);
+      const jobSubmissionResponse = await this.#submitPrompt(body, path);
       const jobStatusResponse = await this.#pollJobStatus(jobSubmissionResponse.job_id, path);
       this.#logDuration('Firefall API Capability Execution call', startTime);
 

diff --git a/packages/spacecat-shared-gpt-client/test/clients/firefall-client.test.js b/packages/spacecat-shared-gpt-client/test/clients/firefall-client.test.js
@@ -38,7 +38,7 @@ describe('FirefallClient', () => {
     mockContext = {
       log: mockLog.object,
       env: {
-        FIREFALL_API_ENDPOINT: 'https://api.firefall.example.com',
+        FIREFALL_API_ENDPOINT: FirefallClient.STAGE_FIREFALL_API_ENDPOINT,
         FIREFALL_API_KEY: 'apiKeyExample',
         FIREFALL_API_POLL_INTERVAL: 100,
         FIREFALL_API_CAPABILITY_NAME: 'gpt4_32k_completions_capability',
@@ -76,7 +76,8 @@ describe('FirefallClient', () => {
     });
   });
 
-  describe('fetchCapabilityExecution', () => {
+  // eslint-disable-next-line func-names
+  describe('fetchCapabilityExecution', function () {
     this.timeout(3000);
     let client;
 
@@ -177,7 +178,8 @@ describe('FirefallClient', () => {
     });
   });
 
-  describe('fetchChatCompletion', () => {
+  // eslint-disable-next-line func-names
+  describe('fetchChatCompletion', function () {
     const chatPath = '/v2/chat/completions';
     const chatResponse = {
       choices: [
@@ -231,34 +233,29 @@ describe('FirefallClient', () => {
         .post(chatPath)
         .reply(200, chatResponseDup);
       const imageUrl = 'iVBORw0KGgoAAAANSUhEUgAAABgAAAAYCAYA...=';
+      const options = {
+        imageUrls: [imageUrl],
+        model: 'gpt-4-turbo',
+        responseFormat: 'json_object',
+      };
 
-      mockLog.expects('warn').once().withArgs('Image URLs were provided but capability (gpt4_32k_completions_capability) may not handle vision prompts. Continuing...');
-      await expect(client.fetchChatCompletion('Test prompt', { imageUrls: [imageUrl] }))
+      await expect(client.fetchChatCompletion('Test prompt', options))
         .to.be.rejectedWith('Invalid response format.');
-      mockLog.verify();
     });
 
     it('should handle a missing response message', async () => {
-      // Run this with a different capability.
-      mockContext.env.FIREFALL_API_CAPABILITY_NAME = 'gpt-4-vision';
-      const clientCapacity = FirefallClient.createFrom(mockContext);
-      const logSpy = sinon.spy(mockContext.log, 'warn');
-
       const chatResponseDup = JSON.parse(JSON.stringify(chatResponse));
       delete chatResponseDup.choices[0].message;
       nock(mockContext.env.FIREFALL_API_ENDPOINT)
         .post(chatPath)
         .reply(200, chatResponseDup);
       const imageUrl = 'iVBORw0KGgoAAAANSUhEUgAAABgAAAAYCAYA...=';
 
-      await expect(clientCapacity.fetchChatCompletion(
+      await expect(client.fetchChatCompletion(
         'Test prompt',
-        { imageUrls: [imageUrl] },
+        { imageUrls: [imageUrl], model: 'gpt-4-vision', responseFormat: 'ignored' },
       ))
         .to.be.rejectedWith('Invalid response format.');
-
-      // Modal (capacity) handles vision prompts, so no warning should be logged.
-      expect(logSpy.callCount).to.equal(0);
     });
 
     it('should handle a missing response content', async () => {