Update main.py

main.py

@@ -15,7 +15,7 @@ from langchain.schema import StrOutputParser
 from langchain.chains.qa_with_sources import load_qa_with_sources_chain
 from langchain.vectorstores import Pinecone
 import pinecone
-from langchain.memory import ChatMessageHistory, ConversationBufferMemory
+from langchain.memory import ChatMessageHistory, ConversationBufferMemory, ConversationSummaryBufferMemory
 import pandas as pd
 import numpy as np
 from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
@@ -95,7 +95,7 @@ def retriever_to_cache():
     vectorstore = Pinecone.from_existing_index(
         index_name=index_name, embedding=embeddings
     )
-    retriever = vectorstore.as_retriever(search_type="similarity_score_threshold", search_kwargs={"score_threshold": .7, "k": 30,"filter": {'categorie': {'$eq': 'OF'}}})
+    retriever = vectorstore.as_retriever(search_type="similarity_score_threshold", search_kwargs={"score_threshold": .7, "k": 30,"filter": {'categorie': {'$eq': 'OF'}}, "include_metadata": True})
     return retriever
 
 @cl.set_chat_profiles
@@ -154,14 +154,15 @@ async def start():
         CONDENSE_QUESTION_PROMPT = PromptTemplate.from_template(_template)
         
         ########## Chain with streaming ##########
-        message_history = ChatMessageHistory()
-        memory = ConversationBufferMemory(
-            memory_key="chat_history",
-            output_key="answer",
-            chat_memory=message_history,
-            return_messages=True,
-        )
-        #llm = ChatAnthropic()
+        #message_history = ChatMessageHistory()
+        #memory = ConversationBufferMemory(
+        #    memory_key="chat_history",
+        #    output_key="answer",
+        #    chat_memory=message_history,
+        #    return_messages=True,
+        #)
+        #llm = ChatAnthropic(temperature=0)
+        
         streaming_llm = ChatAnthropic(
             streaming=True,
             temperature=1,
@@ -170,6 +171,12 @@ async def start():
         #question_generator = LLMChain(llm=llm, prompt=CONDENSE_QUESTION_PROMPT)
         #doc_chain = load_qa_chain(streaming_llm, chain_type="stuff")
         #relevant=retriever_to_cache()
+        memory = ConversationSummaryBufferMemory(
+            llm=streaming_llm,
+            output_key='answer',
+            memory_key='chat_history',
+            return_messages=True
+        )
         qa = ConversationalRetrievalChain.from_llm(
             streaming_llm,
             chain_type="stuff",
@@ -178,6 +185,8 @@ async def start():
             #question_generator=question_generator,
             memory=memory,
             return_source_documents=True,
+            get_chat_history=lambda h : h,
+            verbose=False
         )
     
         cl.user_session.set("conversation_chain", qa)