more

app.py

@@ -46,6 +46,9 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
     )
     response = tokenizer.decode(generated_tokens[0], skip_special_tokens=True)
 
+    # Clean the response to remove any unwanted system content (if any)
+    response = response.replace("system", "").strip()
+
     # Update history
     if history is None:
         history = []
@@ -53,6 +56,7 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
 
     print("history:")
     print(history)
+    
     # Format history for Gradio (strictly enforce the role-content format)
     formatted_history = []
     for entry in history:
@@ -65,6 +69,7 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
     return formatted_history
 
 
+
 # Define the Gradio interface
 demo = gr.ChatInterface(
     fn=respond,