Hi @<1523704667563888640:profile|CooperativeOtter46> , this CLI actually uses Dataset.get_mutable_local_copy() which uses parallel workers (by default, the number of CPU cores, I think).