Can I train a Catboost multioutput model on GPU with early stopping?

Question

I can train a multioutput model with Catboost using the "MultiRMSE" loss_function and eval_metric parameters in CatbootRegressor. The problem is that in this way I