[QUESTION] --use-distributed-optimizer and --data-parallel-sharding-strategy

[bugm]

Hello,
To my knowledge， the argument --use-distributed-optimizer is a implementation of Zero1, I want to know if --data-parallel-sharding-strategy 'optim' is also functionally equivalent Zero1 but in a different way of implementation? Thanks!

[Skylion007]

Pretty sure you need --use-custom-fsdp or --torch-fsdp-2 to enable --data-parallel-sharding-strategy