Update app.py

app.py

@@ -40,7 +40,8 @@ warnings.filterwarnings("ignore", module="langchain")
 # Initialize and set the cache
 set_llm_cache(InMemoryCache())
 
-
+#model='gpt-3.5-turbo'
+model='gpt-4o-mini'
 
 #index_name ="radardata11122024"
 #index_name="radarclintcountrymusic11152024"
@@ -48,14 +49,14 @@ index_name="radarmasterdataset11252024"
 
 embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
 def initialize_gpt_model():
-    return ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o')
+    return ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model=model)
 
 gpt_model = initialize_gpt_model()
 
 
 gpt_embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
 gpt_vectorstore = PineconeVectorStore(index_name=index_name, embedding=gpt_embeddings)
-gpt_retriever = gpt_vectorstore.as_retriever(search_kwargs={'k': 2})
+gpt_retriever = gpt_vectorstore.as_retriever(search_kwargs={'k': 1})
 
 
 
@@ -67,16 +68,18 @@ pc = Pinecone(api_key=os.environ['PINECONE_API_KEY'])
 
 
 vectorstore = PineconeVectorStore(index_name=index_name, embedding=embeddings)
-retriever = vectorstore.as_retriever(search_kwargs={'k': 2})
+retriever = vectorstore.as_retriever(search_kwargs={'k': 1})
 
-chat_model = ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o')
+chat_model = ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model=model)
 
 #code for history
-conversational_memory = ConversationBufferWindowMemory(
-        memory_key='chat_history',
-        k=10,
-        return_messages=True
-    )
+memory = ConversationBufferWindowMemory(k=3)
+conversation=ConversationChain(
+    llm=chat_model,
+    memory=memory,
+    verbose=True
+)        
+
 
 # template =f"""Hello there! As your friendly and knowledgeable guide here in Birmingham, Alabama.Give the short ,precise,crisp and straight-foreward response of maximum 2 sentences and dont greet. 
 # {{context}}
@@ -111,7 +114,6 @@ conversational_memory = ConversationBufferWindowMemory(
 
 template = f"""You’re Annie, a charming media personality created by Amit Lamba, guiding people around Birmingham, Alabama. Like Rick Steves or Anthony Bourdain, you’re here to give friendly, concise, and accurate advice with a dash of humor. 
 Keep answers relevant, upbeat, and engaging, spotlighting partner businesses whenever possible. Be conversational, like chatting with a clever friend, and encourage follow-up questions to create a smooth flow. Make users smile and deliver a delightful, informative experience—like a perfect dessert after a great meal.
-
 {{context}}
 Question: {{question}}
 Helpful Answer:"""
@@ -175,20 +177,29 @@ def generate_audio_elevenlabs(text):
 
 import time
 
-# Main function to handle mode selection with character-by-character streaming
 def handle_mode_selection(mode, chat_history, question):
     if mode == "Normal Chatbot":
-        chat_history.append((question, ""))  # Append user question with an empty response initially
+        # Use memory to store history
+        memory.save_context({"input": question}, {"output": ""})
+        chat_history.append((question, ""))  # Add user's question
 
-        # Get response from Pinecone using the qa_chain
-        response = qa_chain.invoke({"query": question, "context": ""})
+        # Get the context from memory
+        context = memory.load_memory_variables({}).get("history", "")
+
+        # Use QA chain to get the response
+        response = qa_chain.invoke({"query": question, "context": context})
         response_text = response['result']
 
-        # Stream each character in the response text to the chat history
+        # Update memory with the bot's response
+        memory.save_context({"input": question}, {"output": response_text})
+
+        # Stream each character in the response text
         for i, char in enumerate(response_text):
-            chat_history[-1] = (question, chat_history[-1][1] + char)  # Update the last message
-            yield chat_history, "", None  # Yield updated chat history
-            time.sleep(0.05)  # Small delay to simulate streaming
+            chat_history[-1] = (question, chat_history[-1][1] + char)
+            yield chat_history, "", None
+            time.sleep(0.05)  # Simulate streaming
+
+        yield chat_history, "", None
 
     elif mode == "Voice to Voice Conversation":
         response_text = qa_chain({"query": question, "context": ""})['result']
@@ -197,7 +208,6 @@ def handle_mode_selection(mode, chat_history, question):
 
 
 
-
 # Function to add a user's message to the chat history and clear the input box
 def add_message(history, message):
     if message.strip():
@@ -389,4 +399,4 @@ with gr.Blocks(theme="rawrsor1/Everforest") as demo:
     )
 
 # Launch the Gradio interface
-demo.launch(show_error=True)
+demo.launch(show_error=True)