Update app.py

app.py

@@ -32,6 +32,7 @@ from pathlib import Path
 import torchaudio
 import numpy as np
 from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+from langchain_huggingface import HuggingFaceEmbeddings
 
 
 # Neo4j imports
@@ -84,7 +85,7 @@ logging.basicConfig(level=logging.DEBUG)
 
 
 # embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
-embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
+embeddings= HuggingFaceEmbeddings(model="sentence-transformers/all-mpnet-base-v2")
 
 
 #Initialization
@@ -121,12 +122,12 @@ gpt4o_mini_model = initialize_gpt4o_mini_model()
 
 # Existing embeddings and vector store for GPT-4o
 gpt_embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
-gpt_vectorstore = PineconeVectorStore(index_name="italyv109102024", embedding=gpt_embeddings)
+gpt_vectorstore = PineconeVectorStore(index_name="italyv109102024", embedding=embeddings)
 gpt_retriever = gpt_vectorstore.as_retriever(search_kwargs={'k': 5})
 
 # New vector store setup for Phi-3.5
-phi_embeddings = OpenAIEmbeddings(api_key=os.environ['OPENAI_API_KEY'])
-phi_vectorstore = PineconeVectorStore(index_name="italyv109102024", embedding=phi_embeddings)
+phi_embeddings = embeddings_phi
+phi_vectorstore = PineconeVectorStore(index_name="italyv109102024", embedding=embeddings)
 phi_retriever = phi_vectorstore.as_retriever(search_kwargs={'k': 5})
 
 
@@ -340,13 +341,12 @@ Ti prego di fornire i dettagli riguardanti il documento che sto per condividere,
 Ecco i dettagli del documento da considerare:
 - Nome del documento:
 - Pagina:
-- Altre informazioni necessarie:.
-<|end|>
+- Altre informazioni necessarie:.<|end|>
 <|user|>
 {context}
 Question: {question}<|end|>
 <|assistant|>
-Sure! Here's the information :
+Sure! Here's the information:
 """
 
 
@@ -463,129 +463,153 @@ Detailed Answer:
 
 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 and model
-        if choice == "Details" and selected_model == chat_model1:  # GPT-4o-mini
-            prompt_template = PromptTemplate(input_variables=["context", "question"], template=gpt4o_mini_template_details)
-        elif 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'])
+# 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 and model
+#         if choice == "Details" and selected_model == chat_model1:  # GPT-4o-mini
+#             prompt_template = PromptTemplate(input_variables=["context", "question"], template=gpt4o_mini_template_details)
+#         elif 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 == chat_model1:
-                logging.debug("Selected model: LM-3")
-                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_model1,
-                    chain_type="stuff",
-                    retriever=retriever,
-                    chain_type_kwargs={"prompt": prompt_template}
-                )
-                response = qa_chain({"query": message})
-                logging.debug(f"LM-3 response: {response}")
-                return response['result'], extract_addresses(response['result']) 
+#             elif selected_model == chat_model1:
+#                 logging.debug("Selected model: LM-3")
+#                 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_model1,
+#                     chain_type="stuff",
+#                     retriever=retriever,
+#                     chain_type_kwargs={"prompt": prompt_template}
+#                 )
+#                 response = qa_chain({"query": message})
+#                 logging.debug(f"LM-3 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.", []
+#             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.", []
 
-    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):
+    # Logic for Phi-3.5
+    if selected_model == phi_pipe:  # LM-2 Phi-3.5 selected
+        retriever = phi_retriever
+        context_documents = retriever.get_relevant_documents(message)
+        context = "\n".join([doc.page_content for doc in context_documents])
+
+        # Use the correct template for Phi-3.5
+        prompt = phi_custom_template.format(context=context, question=message)
+        
+        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']
+            cleaned_response = clean_response(generated_text)
+            return cleaned_response, extract_addresses(cleaned_response)
+        else:
+            return "No response generated.", []
 
 
 
@@ -1071,16 +1095,32 @@ def handle_retrieval_mode_change(choice):
 
 
 
+# def handle_model_choice_change(selected_model):
+#     if selected_model == "LM-2":
+#         # Disable retrieval mode and select style when LM-2 is selected
+#         return gr.update(interactive=False), gr.update(interactive=False), gr.update(interactive=False)
+#     elif selected_model == "LM-1":
+#         # Enable retrieval mode and select style for LM-1
+#         return gr.update(interactive=True), gr.update(interactive=True), gr.update(interactive=True)
+#     else:
+#         # Default case: allow interaction
+#         return gr.update(interactive=True), gr.update(interactive=True), gr.update(interactive=True)
+
 def handle_model_choice_change(selected_model):
-    if selected_model == "LM-2":
+    if selected_model == "LM-2":  # When LM-2 (Phi-3.5) is selected
         # Disable retrieval mode and select style when LM-2 is selected
-        return gr.update(interactive=False), gr.update(interactive=False), gr.update(interactive=False)
-    elif selected_model == "LM-1":
-        # Enable retrieval mode and select style for LM-1
-        return gr.update(interactive=True), gr.update(interactive=True), gr.update(interactive=True)
+        return (
+            gr.update(interactive=False),  # Disable retrieval mode
+            gr.update(interactive=False),  # Disable style (Details/Conversational)
+            gr.update(interactive=False)   # Disable the model choice itself
+        )
     else:
-        # Default case: allow interaction
-        return gr.update(interactive=True), gr.update(interactive=True), gr.update(interactive=True)
+        # Disable GPT-4o, GPT-4o-mini, and KGF, only Phi-3.5 works
+        return (
+            gr.update(interactive=True),  # Allow retrieval mode for other models
+            gr.update(interactive=True),  # Allow style options for other models
+            gr.update(interactive=True)   # Allow other models to be selected
+        )
 
 #Flux Coding
 
@@ -1340,9 +1380,9 @@ with gr.Blocks(theme='Pijush2023/scikit-learn-pijush') as demo:
             state = gr.State()
 
             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", "LM-3"], value="LM-1")
+            choice = gr.Radio(label="Select Style", choices=["Details", "Conversational"], value="Conversational",interactive=False,visible=False)
+            retrieval_mode = gr.Radio(label="Retrieval Mode", choices=["VDB", "KGF"], value="VDB",interactive=False,visible=False)
+            model_choice = gr.Dropdown(label="Choose Model", choices=["LM-2"], value="LM-2")
 
             # 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])