use dap in openfold

[liuxm117]

hi, I see how to use dap is described in the readme, as follows,

from fastfold.distributed import init_dap

torch.distributed.init_process_group(backend='nccl', init_method='env://')
init_dap(args.dap_size)

I want to know is it possible to use dap instead of deepspeed in openfold

[Shenggan]

OpenFold uses DeepSpeed mainly for its activation_checkpointing, bfloat16 training and ZeRO optimizer. DAP in FastFold is for model parallelism, which is not the same as the function provided by DeepSpeed. In principle, it is possible to use both FastFold's DAP and DeepSpeed in OpenFold.