more

app.py

@@ -16,19 +16,19 @@ FastLanguageModel.for_inference(model)  # Enable optimized inference
 
 # Define the response function
 def respond(message, history, system_message, max_tokens, temperature, top_p):
-    # Add the system message and include previous conversation history
+    # Prepare the messages, separating the system message from user/assistant pairs
     messages = [{"role": "system", "content": system_message}]
     
-    # Append the previous conversation to the message context
+    # Append the conversation history (user-assistant pairs)
     if history:
         for entry in history:
             messages.append({"role": "user", "content": entry["user"]})
             messages.append({"role": "assistant", "content": entry["assistant"]})
 
-    # Add the user's new input
+    # Add the user's new message to the list of messages
     messages.append({"role": "user", "content": message})
 
-    # Tokenize inputs
+    # Tokenize the input
     inputs = tokenizer.apply_chat_template(
         messages,
         tokenize=True,
@@ -48,20 +48,21 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
     )
     response = tokenizer.decode(generated_tokens[0], skip_special_tokens=True)
 
-    # Debug: Print the raw assistant response
-    print("Raw Assistant Response:", response)
-
-    # Clean up the assistant response (strip out unwanted system info)
+    # Clean the response to ensure no system messages are included
     response = response.replace("Cutting Knowledge Date", "").replace("You are a helpful assistant.", "").strip()
 
-    # Update history with the new interaction
+    # Debug: Print the raw and cleaned assistant response
+    print("Raw Assistant Response:", response)
+
+    # Update the conversation history with the new user-assistant interaction
     if history is None:
         history = []
     history.append({"user": message, "assistant": response})
 
+    # Debug: Print updated history
     print("Updated History:", history)
     
-    # Format history for Gradio (enforcing role-content format)
+    # Format the history into the structure expected by Gradio
     formatted_history = []
     for entry in history:
         formatted_history.append({"role": "user", "content": entry["user"]})