constantinpape/torch-em

Request for DistributedDataParallel

Opened this issue · 1 comments

sladenheim commented

To allow for the training of larger models split across GPUS using the DefaultTrainer class, implement model parallel capabilities using the torch.DistributedDataParallel functionality.

constantinpape commented

Thanks for opening this @sladenheim. We are currently focusing on a new release of micro_sam, but will then look into this.