INNER CODE UNIT · Python

main

Guitaricet/relora · torchrun_main.py:338

def main(args):
    # seed all
    torch.manual_seed(args.seed)
    np.random.seed(args.seed)
    random.seed(args.seed)

    assert "LOCAL_RANK" in os.environ, "torchrun should set LOCAL_RANK"
    global_rank = int(os.environ['RANK'])
    local_rank = int(os.environ["LOCAL_RANK"])
    world_size = int(os.environ["WORLD_SIZE"])
    torch.cuda.set_device(local_rank)

    logger.info(f"Global rank {global_rank}, local rank {local_rank}, device: {torch.cuda.current_device()}")

    dist.init_process_group(backend="nccl", rank=global_rank, world_size=world_size)

    logger.info("Process group initialized")
    device = f"cuda:{local_rank}"

View source record →

📰 Research Paper
Loading…
⏳ Fetching content…