remove torch

app.py

@@ -1,7 +1,5 @@
 import re
 
-import torch
-
 import gradio as gr
 
 from peft import PeftModel
@@ -86,8 +84,7 @@ def response(user_schemas, user_question):
                                            return_dict=True,
                                            return_tensors="pt")
     
-    with torch.inference_mode():
-        outputs = model.generate(**inputs, max_new_tokens=1024)
+    outputs = model.generate(**inputs, max_new_tokens=1024)
     
     outputs = tokenizer.batch_decode(outputs)
     output = outputs[0].split("<|im_start|>assistant")[-1]