mnist_ddp.py

#########################################################################
# Reference: https://blog.csdn.net/u010900574/article/details/122780585 #
#########################################################################

import argparse
import os
import time

import torch
import torch.distributed as dist
import torch.nn.functional as F
import torch.optim as optim
from torch.optim.lr_scheduler import StepLR
from torchvision import datasets, transforms

from net import Net


def init_distributed_mode(args):
    """
    Initialize DDP
    """
    os.environ['OMP_NUM_THREADS'] = "1"
    if "RANK" in os.environ and "WORLD_SIZE" in os.environ:
        args.rank = int(os.environ["RANK"])
        args.world_size = int(os.environ["WORLD_SIZE"])
        args.gpu = int(os.environ["LOCAL_RANK"])
    elif "SLURM_PROCID" in os.environ:
        args.rank = int(os.environ["SLURM_PROCID"])
        args.gpu = args.rank % torch.cuda.device_count()
    elif hasattr(args, "rank"):
        pass
    else:
        print("Not using distributed mode")
        args.distributed = False
        return

    args.distributed = True

    torch.cuda.set_device(args.gpu)
    args.dist_backend = "nccl"
    print(
        f"| distributed init (rank {args.rank}): {args.dist_url}, local rank:{args.gpu}, world size:{args.world_size}",
        flush=True)
    dist.init_process_group(
        backend=args.dist_backend, init_method=args.dist_url, world_size=args.world_size, rank=args.rank
    )


def train(args, model, device, train_loader, optimizer, epoch):
    model.train()
    for batch_idx, (data, target) in enumerate(train_loader):
        data, target = data.to(device), target.to(device)
        optimizer.zero_grad()
        output = model(data)
        loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        if args.distributed:
            if dist.get_rank() == 0:
                if batch_idx % args.log_interval == 0:
                    print(
                        "Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}".format(
                            epoch,
                            dist.get_world_size() * batch_idx * len(data),
                            len(train_loader.dataset),
                            100.0 * batch_idx / len(train_loader),
                            loss.item(),
                        )
                    )
        else:
            if batch_idx % args.log_interval == 0:
                print(
                    "Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}".format(
                        epoch,
                        batch_idx * len(data),
                        len(train_loader.dataset),
                        100.0 * batch_idx / len(train_loader),
                        loss.item(),
                    )
                )
        if args.dry_run:
            break


def test(model, device, test_loader):
    model.eval()
    test_loss = 0
    correct = 0
    with torch.no_grad():
        for data, target in test_loader:
            data, target = data.to(device), target.to(device)
            output = model(data)
            test_loss += F.nll_loss(output, target, reduction='sum').item()  # sum up batch loss
            pred = output.argmax(dim=1, keepdim=True)  # get the index of the max log-probability
            correct += pred.eq(target.view_as(pred)).sum().item()

    test_loss /= len(test_loader.dataset)

    print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
        test_loss, correct, len(test_loader.dataset),
        100. * correct / len(test_loader.dataset)))


def main():
    # Training settings
    parser = argparse.ArgumentParser(description='PyTorch MNIST Example')
    parser.add_argument('--batch-size', type=int, default=64, metavar='N',
                        help='input batch size for training (default: 64)')
    parser.add_argument('--test-batch-size', type=int, default=1000, metavar='N',
                        help='input batch size for testing (default: 1000)')
    parser.add_argument('--epochs', type=int, default=14, metavar='N',
                        help='number of epochs to train (default: 14)')
    parser.add_argument('--lr', type=float, default=1.0, metavar='LR',
                        help='learning rate (default: 1.0)')
    parser.add_argument('--gamma', type=float, default=0.7, metavar='M',
                        help='Learning rate step gamma (default: 0.7)')
    parser.add_argument('--no-cuda', action='store_true', default=False,
                        help='disables CUDA training')
    parser.add_argument('--dry-run', action='store_true', default=False,
                        help='quickly check a single pass')
    parser.add_argument('--seed', type=int, default=1, metavar='S',
                        help='random seed (default: 1)')
    parser.add_argument('--log-interval', type=int, default=10, metavar='N',
                        help='how many batches to wait before logging training status')
    parser.add_argument('--save-model', action='store_true', default=False,
                        help='For Saving the current Model')

    parser.add_argument('--local_rank', type=int, help='local rank, will passed by ddp')
    parser.add_argument("--world-size", default=1, type=int, help="number of distributed processes")
    parser.add_argument("--dist-url", default="env://", type=str, help="url used to set up distributed training")
    args = parser.parse_args()
    use_cuda = not args.no_cuda and torch.cuda.is_available()

    init_distributed_mode(args)

    torch.manual_seed(args.seed)

    device = torch.device("cuda" if use_cuda else "cpu")

    train_kwargs = {'batch_size': args.batch_size}
    test_kwargs = {'batch_size': args.test_batch_size}
    if use_cuda:
        cuda_kwargs = {'num_workers': 1,
                       'pin_memory': True,
                       }
        train_kwargs.update(cuda_kwargs)
        test_kwargs.update(cuda_kwargs)

    transform = transforms.Compose([
        transforms.ToTensor(),
        transforms.Normalize((0.1307,), (0.3081,))
    ])

    if args.distributed:
        if torch.distributed.get_rank() != 0:
            # might be downloading mnist data, let rank 0 download first
            torch.distributed.barrier()

    train_dataset = datasets.MNIST('./data', train=True, download=True, transform=transform)

    if args.distributed:
        if torch.distributed.get_rank() == 0:
            # mnist data is downloaded, indicate other ranks can proceed
            torch.distributed.barrier()

    val_dataset = datasets.MNIST('./data', train=False, transform=transform)
    if args.distributed:
        train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset, shuffle=True)
    else:
        train_sampler = torch.utils.data.RandomSampler(train_dataset)
    test_sampler = torch.utils.data.SequentialSampler(val_dataset)

    train_loader = torch.utils.data.DataLoader(train_dataset, sampler=train_sampler, **train_kwargs)
    test_loader = torch.utils.data.DataLoader(val_dataset, sampler=test_sampler, **test_kwargs)

    model = Net().to(device)
    model_without_ddp = model
    if args.distributed:
        model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
        model_without_ddp = model.module

    optimizer = optim.Adadelta(model.parameters(), lr=args.lr)

    scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)

    total_time = 0.

    for epoch in range(1, args.epochs + 1):
        if args.distributed:
            train_sampler.set_epoch(epoch)
        start = time.time()
        train(args, model, device, train_loader, optimizer, epoch)
        total_time += time.time() - start
        if args.distributed:
            # Only run validation on GPU 0 process, for simplicity, so we do not run validation on multi gpu.
            if dist.get_rank() == 0:
                test(model_without_ddp, device, test_loader)
                torch.distributed.barrier()
            else:
                torch.distributed.barrier()
        else:
            test(model, device, test_loader)
        scheduler.step()

    if args.save_model:
        if args.distributed:
            if dist.get_rank() == 0:
                # only save model on GPU0 process.
                torch.save(model.state_dict(), f"mnist_cnn.pt")
        else:
            torch.save(model.state_dict(), f"mnist_cnn_.pt")

    if args.distributed:
        return dist.get_rank(), total_time
    else:
        return 0, total_time


if __name__ == '__main__':
    rk, tt = main()
    print(f'[{rk}] Total time elapsed: {tt} seconds')