use local embedings

clients/groq_client.py

@@ -4,8 +4,8 @@ from uuid import uuid4
 from groq import Groq
 from langchain_core.documents import Document
 
-# CHANGED: Using HuggingFaceEndpointEmbeddings for online API inference
-from langchain_huggingface import HuggingFaceEndpointEmbeddings
+# CHANGED: Using local HuggingFaceEmbeddings instead of online API
+from langchain_huggingface import HuggingFaceEmbeddings
 from langchain_chroma import Chroma
 from dotenv import load_dotenv
 import random
@@ -106,19 +106,16 @@ class GroqClient:
             return []
 
     def init_vector_store(self, documents):
-        # 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,
+        # CHANGED: Using local HuggingFaceEmbeddings instead of online API
+        embeddings_model = HuggingFaceEmbeddings(
+            model_name="sentence-transformers/all-MiniLM-L6-v2",
+            model_kwargs={"device": "cpu"},
+            encode_kwargs={"normalize_embeddings": True},
         )
 
         # Clear old data to avoid duplicates
         if os.path.exists(CHROMA_PATH):
-            shutil.rmtree(CHROMA_PATH)
+            shutil.rmtree(CHROMA_PATH, ignore_errors=True)
 
         uuids = [str(uuid4()) for _ in documents]
 
@@ -128,7 +125,6 @@ class GroqClient:
             persist_directory=CHROMA_PATH,
         )
 
-        # CHANGED: This now makes API calls to HuggingFace for embeddings
         vector_store.add_documents(documents=documents, ids=uuids)
         return vector_store
 

requirements.txt

@@ -5,4 +5,5 @@ python-dotenv
 groq
 langchain-core
 langchain-huggingface
-langchain-chroma
+langchain-chroma
+sentence-transformers