Update app.py

app.py

@@ -99,6 +99,16 @@ def initialize_phi_model():
 
 def initialize_gpt_model():
     return ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o')
+    
+
+def initialize_gpt_mini_model():
+    return ChatOpenAI(api_key=os.environ['OPENAI_API_KEY'], temperature=0, model='gpt-4o-mini')
+
+# Initialize the GPT-4o-mini model
+gpt_mini_model = initialize_gpt_mini_model()
+
+
+
 
 # Initialize both models
 phi_pipe = initialize_phi_model()
@@ -334,12 +344,36 @@ Sure! Here's the information you requested:
 """
 
 
+# def generate_bot_response(history, choice, retrieval_mode, model_choice):
+#     if not history:
+#         return
+
+#     # Select the model
+#     selected_model = chat_model if model_choice == "LM-1" else phi_pipe
+
+#     response, addresses = generate_answer(history[-1][0], choice, retrieval_mode, selected_model)
+#     history[-1][1] = ""
+
+#     for character in response:
+#         history[-1][1] += character
+#         yield history  # Stream each character as it is generated
+#         time.sleep(0.05)  # Add a slight delay to simulate streaming
+
+#     yield history  # Final yield with the complete response
+
 def generate_bot_response(history, choice, retrieval_mode, model_choice):
     if not history:
         return
 
-    # Select the model
-    selected_model = chat_model if model_choice == "LM-1" else phi_pipe
+    # Select the model based on the user's choice
+    if model_choice == "LM-1":
+        selected_model = chat_model
+    elif model_choice == "LM-2":
+        selected_model = phi_pipe
+    elif model_choice == "LM-3":
+        selected_model = gpt_mini_model
+    else:
+        selected_model = chat_model  # Fallback to GPT-4o
 
     response, addresses = generate_answer(history[-1][0], choice, retrieval_mode, selected_model)
     history[-1][1] = ""
@@ -353,6 +387,7 @@ def generate_bot_response(history, choice, retrieval_mode, model_choice):
 
 
 
+
 def generate_tts_response(response, tts_choice):
     with concurrent.futures.ThreadPoolExecutor() as executor:
         if tts_choice == "Alpha":
@@ -451,11 +486,112 @@ def clean_response(response_text):
 
 import traceback
 
+# def generate_answer(message, choice, retrieval_mode, selected_model):
+#     logging.debug(f"generate_answer called with choice: {choice}, retrieval_mode: {retrieval_mode}, and selected_model: {selected_model}")
+
+#     # Logic for disabling options for Phi-3.5
+#     if selected_model == "LM-2":
+#         choice = None
+#         retrieval_mode = None
+
+#     try:
+#         # Select the appropriate template based on the choice
+#         if choice == "Details":
+#             prompt_template = QA_CHAIN_PROMPT_1
+#         elif choice == "Conversational":
+#             prompt_template = QA_CHAIN_PROMPT_2
+#         else:
+#             prompt_template = QA_CHAIN_PROMPT_1  # Fallback to template1
+
+#         # Handle hotel-related queries
+#         if "hotel" in message.lower() or "hotels" in message.lower() and "birmingham" in message.lower():
+#             logging.debug("Handling hotel-related query")
+#             response = fetch_google_hotels()
+#             logging.debug(f"Hotel response: {response}")
+#             return response, extract_addresses(response)
+
+#         # Handle restaurant-related queries
+#         if "restaurant" in message.lower() or "restaurants" in message.lower() and "birmingham" in message.lower():
+#             logging.debug("Handling restaurant-related query")
+#             response = fetch_yelp_restaurants()
+#             logging.debug(f"Restaurant response: {response}")
+#             return response, extract_addresses(response)
+
+#         # Handle flight-related queries
+#         if "flight" in message.lower() or "flights" in message.lower() and "birmingham" in message.lower():
+#             logging.debug("Handling flight-related query")
+#             response = fetch_google_flights()
+#             logging.debug(f"Flight response: {response}")
+#             return response, extract_addresses(response)
+
+#         # Retrieval-based response
+#         if retrieval_mode == "VDB":
+#             logging.debug("Using VDB retrieval mode")
+#             if selected_model == chat_model:
+#                 logging.debug("Selected model: LM-1")
+#                 retriever = gpt_retriever
+#                 context = retriever.get_relevant_documents(message)
+#                 logging.debug(f"Retrieved context: {context}")
+
+#                 prompt = prompt_template.format(context=context, question=message)
+#                 logging.debug(f"Generated prompt: {prompt}")
+
+#                 qa_chain = RetrievalQA.from_chain_type(
+#                     llm=chat_model,
+#                     chain_type="stuff",
+#                     retriever=retriever,
+#                     chain_type_kwargs={"prompt": prompt_template}
+#                 )
+#                 response = qa_chain({"query": message})
+#                 logging.debug(f"LM-1 response: {response}")
+#                 return response['result'], extract_addresses(response['result'])
+
+#             elif selected_model == phi_pipe:
+#                 logging.debug("Selected model: LM-2")
+#                 retriever = phi_retriever
+#                 context_documents = retriever.get_relevant_documents(message)
+#                 context = "\n".join([doc.page_content for doc in context_documents])
+#                 logging.debug(f"Retrieved context for LM-2: {context}")
+
+#                 # Use the correct template variable
+#                 prompt = phi_custom_template.format(context=context, question=message)
+#                 logging.debug(f"Generated LM-2 prompt: {prompt}")
+
+#                 response = selected_model(prompt, **{
+#                     "max_new_tokens": 400,
+#                     "return_full_text": True,
+#                     "temperature": 0.7,
+#                     "do_sample": True,
+#                 })
+
+#                 if response:
+#                     generated_text = response[0]['generated_text']
+#                     logging.debug(f"LM-2 Response: {generated_text}")
+#                     cleaned_response = clean_response(generated_text)
+#                     return cleaned_response, extract_addresses(cleaned_response)
+#                 else:
+#                     logging.error("LM-2 did not return any response.")
+#                     return "No response generated.", []
+
+#         elif retrieval_mode == "KGF":
+#             logging.debug("Using KGF retrieval mode")
+#             response = chain_neo4j.invoke({"question": message})
+#             logging.debug(f"KGF response: {response}")
+#             return response, extract_addresses(response)
+#         else:
+#             logging.error("Invalid retrieval mode selected.")
+#             return "Invalid retrieval mode selected.", []
+
+#     except Exception as e:
+#         logging.error(f"Error in generate_answer: {str(e)}")
+#         logging.error(traceback.format_exc())
+#         return "Sorry, I encountered an error while processing your request.", []
+
 def generate_answer(message, choice, retrieval_mode, selected_model):
     logging.debug(f"generate_answer called with choice: {choice}, retrieval_mode: {retrieval_mode}, and selected_model: {selected_model}")
 
     # Logic for disabling options for Phi-3.5
-    if selected_model == "LM-2":
+    if selected_model == phi_pipe:
         choice = None
         retrieval_mode = None
 
@@ -468,21 +604,19 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
         else:
             prompt_template = QA_CHAIN_PROMPT_1  # Fallback to template1
 
-        # Handle hotel-related queries
+        # Handle hotel, restaurant, and flight-related queries as before
         if "hotel" in message.lower() or "hotels" in message.lower() and "birmingham" in message.lower():
             logging.debug("Handling hotel-related query")
             response = fetch_google_hotels()
             logging.debug(f"Hotel response: {response}")
             return response, extract_addresses(response)
 
-        # Handle restaurant-related queries
         if "restaurant" in message.lower() or "restaurants" in message.lower() and "birmingham" in message.lower():
             logging.debug("Handling restaurant-related query")
             response = fetch_yelp_restaurants()
             logging.debug(f"Restaurant response: {response}")
             return response, extract_addresses(response)
 
-        # Handle flight-related queries
         if "flight" in message.lower() or "flights" in message.lower() and "birmingham" in message.lower():
             logging.debug("Handling flight-related query")
             response = fetch_google_flights()
@@ -492,51 +626,22 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
         # Retrieval-based response
         if retrieval_mode == "VDB":
             logging.debug("Using VDB retrieval mode")
-            if selected_model == chat_model:
-                logging.debug("Selected model: LM-1")
-                retriever = gpt_retriever
-                context = retriever.get_relevant_documents(message)
-                logging.debug(f"Retrieved context: {context}")
-
-                prompt = prompt_template.format(context=context, question=message)
-                logging.debug(f"Generated prompt: {prompt}")
-
-                qa_chain = RetrievalQA.from_chain_type(
-                    llm=chat_model,
-                    chain_type="stuff",
-                    retriever=retriever,
-                    chain_type_kwargs={"prompt": prompt_template}
-                )
-                response = qa_chain({"query": message})
-                logging.debug(f"LM-1 response: {response}")
-                return response['result'], extract_addresses(response['result'])
-
-            elif selected_model == phi_pipe:
-                logging.debug("Selected model: LM-2")
-                retriever = phi_retriever
-                context_documents = retriever.get_relevant_documents(message)
-                context = "\n".join([doc.page_content for doc in context_documents])
-                logging.debug(f"Retrieved context for LM-2: {context}")
-
-                # Use the correct template variable
-                prompt = phi_custom_template.format(context=context, question=message)
-                logging.debug(f"Generated LM-2 prompt: {prompt}")
-
-                response = selected_model(prompt, **{
-                    "max_new_tokens": 400,
-                    "return_full_text": True,
-                    "temperature": 0.7,
-                    "do_sample": True,
-                })
-
-                if response:
-                    generated_text = response[0]['generated_text']
-                    logging.debug(f"LM-2 Response: {generated_text}")
-                    cleaned_response = clean_response(generated_text)
-                    return cleaned_response, extract_addresses(cleaned_response)
-                else:
-                    logging.error("LM-2 did not return any response.")
-                    return "No response generated.", []
+            retriever = gpt_retriever  # Use the same retriever for all GPT models
+            context = retriever.get_relevant_documents(message)
+            logging.debug(f"Retrieved context: {context}")
+
+            prompt = prompt_template.format(context=context, question=message)
+            logging.debug(f"Generated prompt: {prompt}")
+
+            qa_chain = RetrievalQA.from_chain_type(
+                llm=selected_model,
+                chain_type="stuff",
+                retriever=retriever,
+                chain_type_kwargs={"prompt": prompt_template}
+            )
+            response = qa_chain({"query": message})
+            logging.debug(f"Response from {selected_model}: {response}")
+            return response['result'], extract_addresses(response['result'])
 
         elif retrieval_mode == "KGF":
             logging.debug("Using KGF retrieval mode")
@@ -1258,7 +1363,7 @@ with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
             chatbot = gr.Chatbot([], elem_id="RADAR:Channel 94.1", bubble_full_width=False)
             choice = gr.Radio(label="Select Style", choices=["Details", "Conversational"], value="Conversational")
             retrieval_mode = gr.Radio(label="Retrieval Mode", choices=["VDB", "KGF"], value="VDB")
-            model_choice = gr.Dropdown(label="Choose Model", choices=["LM-1", "LM-2"], value="LM-1")
+            model_choice = gr.Dropdown(label="Choose Model", choices=["LM-1", "LM-2", "LM-3"], value="LM-1")
 
             # Link the dropdown change to handle_model_choice_change
             model_choice.change(fn=handle_model_choice_change, inputs=model_choice, outputs=[retrieval_mode, choice, choice])