Update app.py

app.py

@@ -19,8 +19,9 @@ def load_llama_model():
         model_id,
         torch_dtype=torch.bfloat16,
         device_map="auto",
-        # offload_folder="offload", 
-    ).to("cuda")
+        offload_folder="offload", 
+    )
+    model.tie_weights() 
     processor = AutoProcessor.from_pretrained(model_id)
 
     return model, processor