Update utils.py

utils.py

@@ -76,7 +76,7 @@ def load_tokenizer_and_model(base_model, load_8bit=False):
         #model.half()  # seems to fix bugs for some users.
 
     model.eval()
-    return tokenizer,model
+    return tokenizer,model, device
 
 # Greedy Search
 def greedy_search(input_ids: torch.Tensor,