ignore keys for now

modeling.py

@@ -135,7 +135,7 @@ class MARModel(PreTrainedModel):
 
         state_dict = torch.load(checkpoint_path, map_location=device)["model_ema"]
         
-        model_architecture.load_state_dict(state_dict)
+        model_architecture.load_state_dict(state_dict, strict=False)
 
         model_architecture.eval()