funasr/bin/train_ds.py
@@ -81,7 +81,13 @@ deepspeed.init_distributed(dist_backend=kwargs.get("backend", "nccl")) elif use_ddp or use_fsdp: logging.info(f"use_ddp: {use_ddp}, use_fsdp: {use_fsdp}") dist.init_process_group(backend=kwargs.get("backend", "nccl"), init_method="env://") dist.init_process_group( backend=kwargs.get("backend", "nccl"), init_method="env://", find_unused_parameters=kwargs.get("train_conf", {}).get( "find_unused_parameters", False ), ) torch.cuda.set_device(local_rank) # rank = dist.get_rank()