Update app.py

app.py

@@ -402,24 +402,105 @@ Answer:
 """
 
 
+# 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:', '', response_text, flags=re.DOTALL).strip()
+#     return cleaned_response
+    
+
+# def generate_answer(message, choice, retrieval_mode, selected_model):
+#     logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
+
+#     try:
+#         if "hotel" in message.lower() or "hotels" in message.lower() and "birmingham" in message.lower():
+#             response = fetch_google_hotels()
+#             return response, extract_addresses(response)
+        
+#         if "restaurant" in message.lower() or "restaurants" in message.lower() and "birmingham" in message.lower():
+#             response = fetch_yelp_restaurants()
+#             return response, extract_addresses(response)
+
+#         if "flight" in message.lower() or "flights" in message.lower() and "birmingham" in message.lower():
+#             response = fetch_google_flights()
+#             return response, extract_addresses(response)
+
+#         if retrieval_mode == "VDB":
+#             if selected_model == chat_model:
+#                 retriever = gpt_retriever
+#                 prompt_template = QA_CHAIN_PROMPT_1 if choice == "Details" else QA_CHAIN_PROMPT_2
+#                 context = retriever.get_relevant_documents(message)
+#                 prompt = prompt_template.format(context=context, question=message)
+
+#                 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})
+#                 return response['result'], extract_addresses(response['result'])
+
+#             elif selected_model == phi_pipe:
+#                 retriever = phi_retriever
+#                 context = retriever.get_relevant_documents(message)
+#                 prompt = phi_short_template.format(context=context, question=message)
+
+#                 logging.debug(f"Phi-3.5 Prompt: {prompt}")
+
+#                 response = selected_model(prompt, **{
+#                     "max_new_tokens": 128,  # Increased to handle longer responses
+#                     "return_full_text": False,
+#                     "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":
+#             response = chain_neo4j.invoke({"question": message})
+#             return response, extract_addresses(response)
+#         else:
+#             return "Invalid retrieval mode selected.", []
+
+#     except Exception as e:
+#         logging.error(f"Error in generate_answer: {e}")
+#         return "Sorry, I encountered an error while processing your request.", []
+
+
 import re
+import time
 
 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:', '', response_text, flags=re.DOTALL).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 generate_answer(message, choice, retrieval_mode, selected_model):
     logging.debug(f"generate_answer called with choice: {choice} and retrieval_mode: {retrieval_mode}")
 
     try:
+        start_time = time.time()
+
         if "hotel" in message.lower() or "hotels" in message.lower() and "birmingham" in message.lower():
             response = fetch_google_hotels()
             return response, extract_addresses(response)
@@ -455,21 +536,43 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
 
                 logging.debug(f"Phi-3.5 Prompt: {prompt}")
 
-                response = selected_model(prompt, **{
-                    "max_new_tokens": 128,  # Increased to handle longer responses
-                    "return_full_text": False,
-                    "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.", []
+                total_response = ""
+                stop_generation = False
+
+                while not stop_generation:
+                    elapsed_time = time.time() - start_time
+                    if elapsed_time > 5:
+                        logging.warning("Stopping generation due to time limit.")
+                        break  # Stop generation if it exceeds 5 seconds
+
+                    response = selected_model(prompt, **{
+                        "max_new_tokens": 256,  # Generate more tokens at a time to reduce iterations
+                        "return_full_text": False,
+                        "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 Partial Response: {generated_text}")
+
+                        # Append the new text to the total response
+                        total_response += " " + generated_text
+
+                        # Check if the generated text is starting to be irrelevant or the response seems complete
+                        if len(generated_text.strip()) < 256 or "Thank you" in generated_text or generated_text.endswith('.'):
+                            stop_generation = True
+
+                        # Update the prompt to continue the context
+                        prompt = generated_text.strip()
+
+                    else:
+                        logging.error("Phi-3.5 did not return any response.")
+                        return "No response generated.", []
+
+                # Final cleanup of the response
+                cleaned_response = clean_response(total_response)
+                return cleaned_response, extract_addresses(cleaned_response)
 
         elif retrieval_mode == "KGF":
             response = chain_neo4j.invoke({"question": message})
@@ -488,8 +591,6 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
 
 
 
-
-
 def bot(history, choice, tts_choice, retrieval_mode, model_choice):
     if not history:
         return history