Update app.py

app.py

@@ -11,6 +11,9 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
 
     prompt = system_message.strip() + "\n\n"
 
+    # Limit history (VERY important)
+    history = history[-4:]
+
     for user_msg, bot_msg in history:
         prompt += f"User: {user_msg}\nAssistant: {bot_msg}\n"
 
@@ -18,13 +21,15 @@ def respond(message, history, system_message, max_tokens, temperature, top_p):
 
     result = pipe(
         prompt,
-        max_new_tokens=max_tokens,
+        max_new_tokens=min(max_tokens, 4096),  # prevent runaway generation
         temperature=temperature,
         top_p=top_p,
-        do_sample=True
+        do_sample=True,
+        return_full_text=False,
+        eos_token_id=pipe.tokenizer.eos_token_id,  # important stop signal
     )
 
-    return result[0]["generated_text"][len(prompt):]
+    return result[0]["generated_text"]
 
 
 chatbot = gr.ChatInterface(