Update app.py

app.py

@@ -116,59 +116,52 @@ def update_chat(message, history):
     history.append({"role": "user", "content": message})
     return history, message, ""
 
-# def get_assistant_response(message, history, max_tokens, temperature, top_p, qa_chain_state_dict):
-#     """
-#     Generate the assistant's response using the QA chain (if available) or fallback to plain chat.
-#     The pre-prompt is always included by concatenating it to the user's new question.
-#     """
-#     qa_chain = qa_chain_state_dict.get("qa_chain")
-    
-#     if qa_chain is not None:
-#         # Format history to the plain-text format expected by the QA chain
-#         formatted_history = format_chat_history(history)
-#         # Prepend the pre-prompt to the current question
-#         combined_question = PRE_PROMPT + "\n" + message
-#         response = qa_chain.invoke({"question": combined_question, "chat_history": formatted_history})
-#         answer = response.get("answer", "")
-#         history.append({"role": "assistant", "content": answer})
-#         return history, {"qa_chain": qa_chain}
-    
-#     # Fallback: Plain Chat Mode using the InferenceClient (pre-prompt already included here)
-#     messages = [{"role": "system", "content": PRE_PROMPT}] + history
-#     response = ""
-#     result = client.chat_completion(
-#         messages,
-#         max_tokens=max_tokens,
-#         stream=False,
-#         temperature=temperature,
-#         top_p=top_p,
-#     )
-#     for token_message in result:
-#         token = token_message.choices[0].delta.content
-#         response += token
-    
-#     history.append({"role": "assistant", "content": response})
-#     return history, {"qa_chain": qa_chain}
-
 def get_assistant_response(message, history, max_tokens, temperature, top_p, qa_chain_state_dict):
     qa_chain = qa_chain_state_dict.get("qa_chain")
     
     if qa_chain is not None:
-        # Format history to the plain-text format expected by the QA chain
+        # Format chat history to the plain-text format expected by the QA chain.
         formatted_history = format_chat_history(history)
-        # Prepend the pre-prompt to the current question
-        combined_question = PRE_PROMPT + "\n" + message
+        
+        # Update the pre-prompt to encourage speculative responses.
+        speculative_pre_prompt = PRE_PROMPT + "\nIf you're not completely sure, please provide your best guess and mention that it is speculative."
+        combined_question = speculative_pre_prompt + "\n" + message
+        
+        # Try retrieving an answer via the QA chain.
         response = qa_chain.invoke({"question": combined_question, "chat_history": formatted_history})
         answer = response.get("answer", "").strip()
         
-        # Check if the answer is empty and apply a fallback response if needed.
+        # If no answer is returned, try the fallback plain chat mode with adjusted parameters.
         if not answer:
-            answer = "I'm sorry, I couldn't retrieve a clear answer. Feel free t ocontact consider Christopher directly: https://gcmarais.com/contact/"
+            # Increase temperature and optionally max_tokens for fallback.
+            increased_temperature = min(temperature + 0.2, 1.0)  # Cap temperature at 1.0
+            increased_max_tokens = max_tokens + 128  # Increase max tokens for a longer response if needed
+            
+            speculative_prompt = speculative_pre_prompt + "\n" + message
+            messages = [{"role": "system", "content": speculative_prompt}] + history
+            response = ""
+            result = client.chat_completion(
+                messages,
+                max_tokens=increased_max_tokens,
+                stream=False,
+                temperature=increased_temperature,
+                top_p=top_p,
+            )
+            for token_message in result:
+                token = token_message.choices[0].delta.content
+                response += token
+            answer = response.strip()
+            
+            # Final fallback if still empty.
+            if not answer:
+                answer = ("I'm sorry, I couldn't retrieve a clear answer. "
+                          "However, based on the available context, here is my best guess: "
+                          "[speculative answer].")
         
         history.append({"role": "assistant", "content": answer})
         return history, {"qa_chain": qa_chain}
     
-    # Fallback: Plain Chat Mode using the InferenceClient
+    # Fallback: Plain Chat Mode using the InferenceClient when no QA chain is available.
     messages = [{"role": "system", "content": PRE_PROMPT}] + history
     response = ""
     result = client.chat_completion(
@@ -184,12 +177,14 @@ def get_assistant_response(message, history, max_tokens, temperature, top_p, qa_
 
     response = response.strip()
     if not response:
-        response = "I'm sorry, I couldn't generate a response. Please try asking in a different way. Alterantively, consider contacting Christopher directly: https://gcmarais.com/contact/"
+        response = ("I'm sorry, I couldn't generate a response. Please try asking in a different way. "
+                    "Alternatively, consider contacting Christopher directly: https://gcmarais.com/contact/")
     
     history.append({"role": "assistant", "content": response})
     return history, {"qa_chain": qa_chain}
 
 
+
 # Global InferenceClient for plain chat (fallback)
 client = InferenceClient("deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B")