Update app.py

app.py

@@ -402,87 +402,6 @@ 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
 
 def clean_response(response_text):
@@ -492,9 +411,10 @@ def clean_response(response_text):
     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()
+    #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}")
@@ -535,36 +455,21 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
 
                 logging.debug(f"Phi-3.5 Prompt: {prompt}")
 
-                # Initialize variables for complete response generation
-                total_response = ""
-                stop_condition = False
-
-                while not stop_condition:
-                    response = selected_model(prompt, **{
-                        "max_new_tokens": 128,  # Generate 128 tokens at a time
-                        "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']
-                        total_response += generated_text
-                        logging.debug(f"Phi-3.5 Partial Response: {generated_text}")
-
-                        # Check if the response seems complete or if we've hit the maximum length
-                        if len(generated_text.strip()) < 128:  # Assuming shorter output indicates completion
-                            stop_condition = True
-                        else:
-                            # Update the prompt with the new context to continue generating
-                            prompt = generated_text.strip()
-
-                    else:
-                        logging.error("Phi-3.5 did not return any response.")
-                        return "No response generated.", []
-
-                cleaned_response = clean_response(total_response)
-                return cleaned_response, extract_addresses(cleaned_response)
+                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})
@@ -582,6 +487,9 @@ def generate_answer(message, choice, retrieval_mode, selected_model):
 
 
 
+
+
+
 def bot(history, choice, tts_choice, retrieval_mode, model_choice):
     if not history:
         return history