Update app.py

app.py

@@ -68,7 +68,7 @@ def query_teapot(prompt, context, user_input):
     inputs = tokenizer(input_text, return_tensors="pt")
     input_length = inputs["input_ids"].shape[1]
     
-    output = model.generate(**inputs, max_new_tokens=max_new_tokens)
+    output = model.generate(**inputs, max_new_tokens=512)
     
     output_text = tokenizer.decode(output[0], skip_special_tokens=True)
     total_length = output.shape[1]  # Includes both input and output tokens