fix

app.py

@@ -61,6 +61,10 @@ def chat_with_model(messages):
     if pad_id is None:
         pad_id = current_tokenizer.unk_token_id or 0
 
+    output_text = ""
+    in_think = False
+    max_new_tokens = 256
+
     prompt = format_prompt(messages)
     device = torch.device("cuda")
     current_model.to(device).half()
@@ -80,7 +84,7 @@ def chat_with_model(messages):
     # 3. Build generation kwargs
     generation_kwargs = dict(
         **inputs,
-        max_new_tokens=256,
+        max_new_tokens=max_new_tokens,
         do_sample=True,
         streamer=streamer,
         eos_token_id=eos_id,