updare index and reranker

.gitignore

@@ -0,0 +1,2 @@
+.env
+*.ipynb

app.py

@@ -4,6 +4,9 @@ monkey.patch_all()
 import nltk
 nltk.download('punkt_tab')
 
+import nltk
+nltk.download('punkt_tab')
+
 import os
 from dotenv import load_dotenv
 import asyncio
@@ -20,10 +23,10 @@ from pinecone import Pinecone
 from pinecone_text.sparse import BM25Encoder
 from langchain_huggingface import HuggingFaceEmbeddings
 from langchain_community.retrievers import PineconeHybridSearchRetriever
-from langchain_groq import ChatGroq
 from langchain.retrievers import ContextualCompressionRetriever
-from langchain.retrievers.document_compressors import FlashrankRerank
 from langchain_community.chat_models import ChatPerplexity
+from langchain.retrievers.document_compressors import CrossEncoderReranker
+from langchain_community.cross_encoders import HuggingFaceCrossEncoder
 
 # Load environment variables
 load_dotenv(".env")
@@ -62,7 +65,7 @@ def initialize_pinecone(index_name: str):
 ##################################################
 
 # Initialize Pinecone index and BM25 encoder
-pinecone_index = initialize_pinecone("updated-mbzuai-policies")
+pinecone_index = initialize_pinecone("updated-mbzuai-policies-17112024")
 bm25 = BM25Encoder().load("./new_mbzuai-policies.json")
 
 ##################################################
@@ -77,7 +80,8 @@ retriever = PineconeHybridSearchRetriever(
     sparse_encoder=bm25, 
     index=pinecone_index, 
     top_k=20, 
-    alpha=0.5
+    alpha=0.5,
+    
 )
 
 # Initialize LLM
@@ -86,7 +90,11 @@ llm = ChatPerplexity(temperature=0, pplx_api_key=GROQ_API_KEY, model="llama-3.1-
 
 
 # Initialize Reranker
-compressor = FlashrankRerank()
+# compressor = FlashrankRerank()
+model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-base")
+compressor = CrossEncoderReranker(model=model, top_n=20)
+
+
 compression_retriever = ContextualCompressionRetriever(
     base_compressor=compressor, base_retriever=retriever
 )
@@ -191,14 +199,21 @@ def handle_message(data):
     else:
         language = "Arabic"
     session_id = data.get('session_id', SESSION_ID_DEFAULT)
-    chain = conversational_rag_chain.pick("answer")
+    # chain = conversational_rag_chain.pick("answer")
+
+    # try:
+    #     for chunk in conversational_rag_chain.stream(
+    #             {"input": question, 'language': language},
+    #             config={"configurable": {"session_id": session_id}},
+    #         ):
+    #         emit('response', chunk, room=request.sid)
+    # except Exception as e:
+    #     print(f"Error during message handling: {e}")
+    #     emit('response', "An error occurred while processing your request." + str(e), room=request.sid)
 
     try:
-        for chunk in chain.stream(
-                {"input": question, 'language': language},
-                config={"configurable": {"session_id": session_id}},
-            ):
-            emit('response', chunk, room=request.sid)
+        response = conversational_rag_chain.invoke({"input": question, 'language': language}, config={"configurable": {"session_id": session_id}})
+        emit('response', response, room=request.sid)
     except Exception as e:
         print(f"Error during message handling: {e}")
         emit('response', "An error occurred while processing your request." + str(e), room=request.sid)