Merge branch 'online-st'

clients/groq_client.py

@@ -4,8 +4,8 @@ from uuid import uuid4
 from groq import Groq
 from langchain_core.documents import Document
 
-# CHANGED: Replaced HuggingFaceEndpointEmbeddings with HuggingFaceEmbeddings for local inference
-from langchain_huggingface import HuggingFaceEmbeddings
+# CHANGED: Using HuggingFaceEndpointEmbeddings for online API inference
+from langchain_huggingface import HuggingFaceEndpointEmbeddings
 from langchain_chroma import Chroma
 from dotenv import load_dotenv
 import random
@@ -22,6 +22,7 @@ load_dotenv(dotenv_path=".secrets.env", override=True)
 
 # Config
 GROQ_API_KEY = os.getenv("GROQ_API_KEY")
+HF_API_KEY = os.getenv("HF_API_KEY")
 DATA_PATH = "data.json"
 CHROMA_PATH = "chroma_db"
 
@@ -105,14 +106,14 @@ class GroqClient:
             return []
 
     def init_vector_store(self, documents):
-        # CHANGED: Replaced online HuggingFaceEndpointEmbeddings with local HuggingFaceEmbeddings
-        # This downloads and stores the embedding model locally, eliminating API dependency
-        embeddings_model = HuggingFaceEmbeddings(
-            model_name="sentence-transformers/all-MiniLM-L6-v2",
-            model_kwargs={"device": "cpu"},  # Force CPU usage to avoid GPU conflicts
-            encode_kwargs={
-                "normalize_embeddings": True
-            },  # Normalize embeddings for better similarity search
+        # CHANGED: Using online HuggingFaceEndpointEmbeddings with API key
+        # This uses HuggingFace's hosted inference API instead of local model
+        if not HF_API_KEY:
+            raise RuntimeError("HF_API_KEY not found in environment")
+
+        embeddings_model = HuggingFaceEndpointEmbeddings(
+            model="sentence-transformers/all-MiniLM-L6-v2",
+            huggingfacehub_api_token=HF_API_KEY,
         )
 
         # Clear old data to avoid duplicates
@@ -127,7 +128,7 @@ class GroqClient:
             persist_directory=CHROMA_PATH,
         )
 
-        # CHANGED: This now processes embeddings locally instead of making API calls
+        # CHANGED: This now makes API calls to HuggingFace for embeddings
         vector_store.add_documents(documents=documents, ids=uuids)
         return vector_store