Update app.py

app.py

@@ -48,14 +48,28 @@ def load_model(model_name):
             del model
             torch.cuda.empty_cache()
         
-        tokenizer = AutoTokenizer.from_pretrained(models[model_name], padding_side='left', token=hugging_face_token, trust_remote_code=True)
+        tokenizer = AutoTokenizer.from_pretrained(
+            models[model_name], 
+            padding_side='left', 
+            use_auth_token=hugging_face_token, 
+            trust_remote_code=True
+        )
         
         # Ensure the padding token is set
         if tokenizer.pad_token is None:
             tokenizer.pad_token = tokenizer.eos_token
             tokenizer.add_special_tokens({'pad_token': tokenizer.eos_token})
         
-        model = AutoModelForCausalLM.from_pretrained(models[model_name], token=hugging_face_token, trust_remote_code=True)
+        model = AutoModelForCausalLM.from_pretrained(
+            models[model_name], 
+            use_auth_token=hugging_face_token, 
+            trust_remote_code=True
+        )
+        
+        # Only move to CUDA if it's not a quantized model
+        if model_name not in quantized_models:
+            model = model.to("cuda")
+        
         selected_model = model_name
     except Exception as e:
         print(f"Error loading model {model_name}: {e}")