adjusting google to do semantic similarity, adjusting config to put api prefixes in front

Author: wjohnsto

services/video-search/src/config.ts

@@ -35,9 +35,9 @@ const {
 } = process.env;
 
 const DEFAULT_VIDEO_INDEX_NAME = 'idx-videos';
-const DEFAULT_VIDEO_PREFIX = 'video';
+const DEFAULT_VIDEO_PREFIX = 'video:';
 const DEFAULT_VECTOR_SET = 'video-vectors';
-const DEFAULT_SUMMARY_PREFIX = 'video-summary';
+const DEFAULT_SUMMARY_PREFIX = 'video-summary:';
 
 export default {
   app: {
@@ -70,35 +70,35 @@ export default {
     API_KEY: SEARCHAPI_API_KEY ?? '',
   },
   hf: {
-    VIDEO_INDEX_NAME: HF_VIDEO_INDEX_NAME ?? `${DEFAULT_VIDEO_INDEX_NAME}-hf`,
-    VIDEO_PREFIX: HF_VIDEO_PREFIX ?? `${DEFAULT_VIDEO_PREFIX}-hf:`,
+    VIDEO_INDEX_NAME: HF_VIDEO_INDEX_NAME ?? `hf-${DEFAULT_VIDEO_INDEX_NAME}`,
+    VIDEO_PREFIX: HF_VIDEO_PREFIX ?? `hf-${DEFAULT_VIDEO_PREFIX}`,
     EMBEDDING_MODEL: HF_EMBEDDING_MODEL ?? 'Xenova/all-MiniLM-L6-v2',
     SUMMARY_MODEL: HF_SUMMARY_MODEL ?? 'Xenova/paraphrase-albert-small-v2',
-    VECTOR_SET: HF_VECTOR_SET ?? `${DEFAULT_VECTOR_SET}-hf`,
-    SUMMARY_PREFIX: HF_SUMMARY_PREFIX ?? `${DEFAULT_SUMMARY_PREFIX}-hf:`,
+    VECTOR_SET: HF_VECTOR_SET ?? `hf-${DEFAULT_VECTOR_SET}`,
+    SUMMARY_PREFIX: HF_SUMMARY_PREFIX ?? `hf-${DEFAULT_SUMMARY_PREFIX}`,
   },
   google: {
     VIDEO_INDEX_NAME:
-      GOOGLE_VIDEO_INDEX_NAME ?? `${DEFAULT_VIDEO_INDEX_NAME}-google`,
-    VIDEO_PREFIX: GOOGLE_VIDEO_PREFIX ?? `${DEFAULT_VIDEO_PREFIX}-google:`,
+      GOOGLE_VIDEO_INDEX_NAME ?? `google-${DEFAULT_VIDEO_INDEX_NAME}`,
+    VIDEO_PREFIX: GOOGLE_VIDEO_PREFIX ?? `google-${DEFAULT_VIDEO_PREFIX}`,
     API_KEY: GOOGLE_API_KEY,
     EMBEDDING_MODEL: GOOGLE_EMBEDDING_MODEL ?? 'embedding-001',
     SUMMARY_MODEL: GOOGLE_SUMMARY_MODEL ?? 'gemini-pro',
-    VECTOR_SET: GOOGLE_VECTOR_SET ?? `${DEFAULT_VECTOR_SET}-google`,
+    VECTOR_SET: GOOGLE_VECTOR_SET ?? `google-${DEFAULT_VECTOR_SET}`,
     SUMMARY_PREFIX:
-      GOOGLE_SUMMARY_PREFIX ?? `${DEFAULT_SUMMARY_PREFIX}-google:`,
+      GOOGLE_SUMMARY_PREFIX ?? `google-${DEFAULT_SUMMARY_PREFIX}`,
   },
   openai: {
     VIDEO_INDEX_NAME:
-      OPENAI_VIDEO_INDEX_NAME ?? `${DEFAULT_VIDEO_INDEX_NAME}-openai`,
-    VIDEO_PREFIX: OPENAI_VIDEO_PREFIX ?? `${DEFAULT_VIDEO_PREFIX}-openai:`,
+      OPENAI_VIDEO_INDEX_NAME ?? `openai-${DEFAULT_VIDEO_INDEX_NAME}`,
+    VIDEO_PREFIX: OPENAI_VIDEO_PREFIX ?? `openai-${DEFAULT_VIDEO_PREFIX}`,
     API_KEY: OPENAI_API_KEY,
     ORGANIZATION: OPENAI_ORGANIZATION,
     EMBEDDING_MODEL: OPENAI_EMBEDDING_MODEL ?? 'gpt-4',
     SUMMARY_MODEL: OPENAI_SUMMARY_MODEL ?? 'gpt-4',
-    VECTOR_SET: OPENAI_VECTOR_SET ?? `${DEFAULT_VECTOR_SET}-openai`,
+    VECTOR_SET: OPENAI_VECTOR_SET ?? `openai-${DEFAULT_VECTOR_SET}`,
     SUMMARY_PREFIX:
-      OPENAI_SUMMARY_PREFIX ?? `${DEFAULT_SUMMARY_PREFIX}-openai:`,
+      OPENAI_SUMMARY_PREFIX ?? `openai-${DEFAULT_SUMMARY_PREFIX}`,
   },
   use: {
     GOOGLE: USE === 'GOOGLE',

services/video-search/src/google/config.ts

@@ -2,17 +2,19 @@ import config from '../config.js';
 import { client } from '../db.js';
 import { RedisVectorStore } from 'langchain/vectorstores/redis';
 import { ChatGoogleGenerativeAI, GoogleGenerativeAIEmbeddings } from '@langchain/google-genai';
+import { TaskType } from '@google/generative-ai';
 
 export const llm = new ChatGoogleGenerativeAI({
   apiKey: config.google.API_KEY,
   modelName: config.google.SUMMARY_MODEL,
-  maxOutputTokens: 10000,
+  maxOutputTokens: 2048,
 });
 
 export function getEmbeddings(modelName?: string) {
   return new GoogleGenerativeAIEmbeddings({
     apiKey: config.google.API_KEY,
-    modelName: modelName ?? config.google.EMBEDDING_MODEL
+    modelName: modelName ?? config.google.EMBEDDING_MODEL,
+    taskType: TaskType.SEMANTIC_SIMILARITY
   });
 }