Update app.py

app.py

@@ -540,33 +540,184 @@ chain_neo4j = (
 
 
 
-import re
+# import re
+
+# def clean_response(response_text):
+#     # Remove any metadata-like information and focus on the main content
+#     # Removes "Document(metadata=...)" and other similar patterns
+#     cleaned_response = re.sub(r'Document\(metadata=.*?\),?\s*', '', response_text, flags=re.DOTALL)
+#     cleaned_response = re.sub(r'page_content=".*?"\),?', '', cleaned_response, flags=re.DOTALL)
+#     cleaned_response = re.sub(r'\[.*?\]', '', cleaned_response, flags=re.DOTALL)  # Remove content in brackets
+#     cleaned_response = re.sub(r'\s+', ' ', cleaned_response).strip()
+#     # Remove any unwanted follow-up questions or unnecessary text
+#     cleaned_response = re.sub(r'Question:.*\nAnswer:', '', cleaned_response, flags=re.DOTALL).strip()
+#     return cleaned_response
+
+
+
+
+# # Define the custom template for Phi-3.5
+# phi_custom_template = """
+# <|system|>
+# You are a helpful assistant.<|end|>
+# <|user|>
+# {context}
+# {question}<|end|>
+# <|assistant|>
+# """
+# 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}")
+
+#     try:
+#         # 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: GPT-4o")
+#                 retriever = gpt_retriever
+#                 prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
+#                 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"GPT-4o response: {response}")
+#                 return response['result'], extract_addresses(response['result'])
+
+#             elif selected_model == phi_pipe:
+#                 logging.debug("Selected model: Phi-3.5")
+#                 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 Phi-3.5: {context}")
+
+#                 # Use the correct template variable
+#                 prompt = phi_custom_template.format(context=context, question=message)
+#                 logging.debug(f"Generated Phi-3.5 prompt: {prompt}")
+
+#                 response = selected_model(prompt, **{
+#                     "max_new_tokens": 160,  # Increased to handle longer responses
+#                     "return_full_text": True,
+#                     "temperature": 0.7,  # Adjusted to avoid cutting off
+#                     "do_sample": True,   # Allow sampling to increase response diversity
+#                 })
+
+#                 if response:
+#                     generated_text = response[0]['generated_text']
+#                     logging.debug(f"Phi-3.5 Response: {generated_text}")
+#                     cleaned_response = clean_response(generated_text)
+#                     return cleaned_response, extract_addresses(cleaned_response)
+#                 else:
+#                     logging.error("Phi-3.5 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 clean_response(response_text):
-    # Remove any metadata-like information and focus on the main content
-    # Removes "Document(metadata=...)" and other similar patterns
-    cleaned_response = re.sub(r'Document\(metadata=.*?\),?\s*', '', response_text, flags=re.DOTALL)
-    cleaned_response = re.sub(r'page_content=".*?"\),?', '', cleaned_response, flags=re.DOTALL)
-    cleaned_response = re.sub(r'\[.*?\]', '', cleaned_response, flags=re.DOTALL)  # Remove content in brackets
-    cleaned_response = re.sub(r'\s+', ' ', cleaned_response).strip()
-    # Remove any unwanted follow-up questions or unnecessary text
-    cleaned_response = re.sub(r'Question:.*\nAnswer:', '', cleaned_response, flags=re.DOTALL).strip()
-    return cleaned_response
 
 
 
+# def bot(history, choice, tts_choice, retrieval_mode, model_choice):
+#     if not history:
+#         return history
+    
+#     # Select the model
+#     selected_model = chat_model if model_choice == "GPT-4o" else phi_pipe
+    
+#     response, addresses = generate_answer(history[-1][0], choice, retrieval_mode, selected_model)
+#     history[-1][1] = ""
+
+#     with concurrent.futures.ThreadPoolExecutor() as executor:
+#         if tts_choice == "Alpha":
+#             audio_future = executor.submit(generate_audio_elevenlabs, response)
+#         elif tts_choice == "Beta":
+#             audio_future = executor.submit(generate_audio_parler_tts, response)
+#         # elif tts_choice == "Gamma":
+#         #     audio_future = executor.submit(generate_audio_mars5, response)
+
+#         for character in response:
+#             history[-1][1] += character
+#             time.sleep(0.05)
+#             yield history, None
+
+#         audio_path = audio_future.result()
+#         yield history, audio_path
+
+#     history.append([response, None])
+
 
-# Define the custom template for Phi-3.5
 phi_custom_template = """
 <|system|>
-You are a helpful assistant.<|end|>
+You are a helpful assistant who provides clear, organized, and conversational responses.<|end|>
 <|user|>
 {context}
-{question}<|end|>
+Question: {question}<|end|>
 <|assistant|>
+Sure! Here's the information you requested:
 """
-import traceback
+import re
+
+def clean_response(response_text):
+    # Remove system and user tags
+    response_text = re.sub(r'<\|system\|>.*?<\|end\|>', '', response_text, flags=re.DOTALL)
+    response_text = re.sub(r'<\|user\|>.*?<\|end\|>', '', response_text, flags=re.DOTALL)
+    response_text = re.sub(r'<\|assistant\|>', '', response_text, flags=re.DOTALL)
+    
+    # Clean up the text by removing extra whitespace
+    cleaned_response = response_text.strip()
+    cleaned_response = re.sub(r'\s+', ' ', cleaned_response)
+    
+    # Ensure the response is conversational and organized
+    cleaned_response = cleaned_response.replace('1.', '\n1.').replace('2.', '\n2.').replace('3.', '\n3.').replace('4.', '\n4.').replace('5.', '\n5.')
+    
+    return cleaned_response
 
+
+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}")
 
@@ -627,10 +778,10 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
                 logging.debug(f"Generated Phi-3.5 prompt: {prompt}")
 
                 response = selected_model(prompt, **{
-                    "max_new_tokens": 160,  # Increased to handle longer responses
+                    "max_new_tokens": 160,
                     "return_full_text": True,
-                    "temperature": 0.7,  # Adjusted to avoid cutting off
-                    "do_sample": True,   # Allow sampling to increase response diversity
+                    "temperature": 0.7,
+                    "do_sample": True,
                 })
 
                 if response:
@@ -660,39 +811,6 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
 
 
 
-
-def bot(history, choice, tts_choice, retrieval_mode, model_choice):
-    if not history:
-        return history
-    
-    # Select the model
-    selected_model = chat_model if model_choice == "GPT-4o" else phi_pipe
-    
-    response, addresses = generate_answer(history[-1][0], choice, retrieval_mode, selected_model)
-    history[-1][1] = ""
-
-    with concurrent.futures.ThreadPoolExecutor() as executor:
-        if tts_choice == "Alpha":
-            audio_future = executor.submit(generate_audio_elevenlabs, response)
-        elif tts_choice == "Beta":
-            audio_future = executor.submit(generate_audio_parler_tts, response)
-        # elif tts_choice == "Gamma":
-        #     audio_future = executor.submit(generate_audio_mars5, response)
-
-        for character in response:
-            history[-1][1] += character
-            time.sleep(0.05)
-            yield history, None
-
-        audio_path = audio_future.result()
-        yield history, audio_path
-
-    history.append([response, None])
-
-
-
-
-
 def add_message(history, message):
     history.append((message, None))
     return history, gr.Textbox(value="", interactive=True, show_label=False)