pytorch.py

from __future__ import print_function
from time import time
import argparse
import os
import sys
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
from torch.optim.lr_scheduler import StepLR
# Distributed Training Releated Imports
from torch.nn.parallel.distributed import DistributedDataParallel
from torch.utils.data import DistributedSampler
import torch.multiprocessing as mp
# Distributed Training Releated Imports
# Profling Related Imports
from torch.profiler import profile, record_function, ProfilerActivity, schedule
# Profling Related Imports


class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 32, 3, 1)
        self.conv2 = nn.Conv2d(32, 64, 3, 1)
        self.dropout1 = nn.Dropout(0.25)
        self.dropout2 = nn.Dropout(0.5)
        self.fc1 = nn.Linear(9216, 128)
        self.fc2 = nn.Linear(128, 10)

    def forward(self, x):
        x = self.conv1(x)
        x = F.relu(x)
        x = self.conv2(x)
        x = F.relu(x)
        x = F.max_pool2d(x, 2)
        x = self.dropout1(x)
        x = torch.flatten(x, 1)
        x = self.fc1(x)
        x = F.relu(x)
        x = self.dropout2(x)
        x = self.fc2(x)
        output = F.log_softmax(x, dim=1)
        return output


def train(args, model, device, train_loader, optimizer, epoch, accumulate_grad_batches):
    model.train()

    my_schedule = schedule(
        skip_first=0,
        wait=1,
        warmup=1,
        active=1,
        repeat=1
    )

    def trace_handler(p):
        output = p.key_averages().table(sort_by="self_cuda_time_total", row_limit=10)
        print(output)
        p.export_chrome_trace("/tmp/trace_" + str(p.step_num) + ".json")

    with profile(
        activities=[ProfilerActivity.CPU],
        with_stack=False,
        schedule=my_schedule,
        on_trace_ready=trace_handler,

    ) as prof:

        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()
            if (batch_idx % accumulate_grad_batches == 0 or batch_idx == len(train_loader) - 1):
                optimizer.step()
            if batch_idx % args.log_interval == 0:
                print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                    epoch, batch_idx * len(data), len(train_loader.dataset),
                    100. * batch_idx / len(train_loader), loss.item()))
                if args.dry_run:
                    break

            prof.step()


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)))


# Create progress group
def setup_ddp(rank, world_size):
    """Setup ddp enviroment"""
    os.environ["MASTER_ADDR"] = "localhost"
    os.environ["MASTER_PORT"] = "8088"
    create_progress_group(rank, world_size)


def create_progress_group(rank, world_size):
    print(f"REGISTERING RANK {rank}")
    if torch.distributed.is_available() and sys.platform not in ("win32", "cygwin"):
        torch.distributed.init_process_group("nccl", rank=rank, world_size=world_size)
# Create progress group


def main(rank, world_size, ddp_spawn):
    t0 = time()
    # 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=3, 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('--use_ddp', type=int, default=1, metavar='N', help='Whether to use DDP')
    parser.add_argument('--accumulate_grad_batches', type=int, default=2, metavar='N', help='How to perform gradient accumulation')
    args = parser.parse_args()
    use_cuda = not args.no_cuda and torch.cuda.is_available()

    if args.use_ddp:
        # Setup DDP
        setup_ddp(rank, world_size)
        torch.cuda.set_device(f"cuda:{rank}")
        # Setup DDP

    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,))
    ])
    dataset1 = datasets.MNIST('../data', train=True, download=True, transform=transform)
    dataset2 = datasets.MNIST('../data', train=False, transform=transform)

    # Create distributed Sampler
    if args.use_ddp:
        train_kwargs['sampler'] = DistributedSampler(dataset1, num_replicas=world_size, rank=rank, shuffle=False)
        test_kwargs['sampler'] = DistributedSampler(dataset2, num_replicas=world_size, rank=rank, shuffle=False)

    train_loader = torch.utils.data.DataLoader(dataset1, **train_kwargs)
    test_loader = torch.utils.data.DataLoader(dataset2, **test_kwargs)
    # Create distributed Sampler

    model = Net().to(device)

    if args.use_ddp:
        # Wrap into DistributedDataParallel
        model = DistributedDataParallel(model, device_ids=[rank])
        # Wrap into DistributedDataParallel

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

    scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
    for epoch in range(1, args.epochs + 1):
        train(args, model, device, train_loader, optimizer, epoch, args.accumulate_grad_batches)
        test(model, device, test_loader)
        scheduler.step()

    # Save only on rank 0 to avoid rank 1 to overrides the checkpoint
    if not args.use_ddp or rank == 0:
        torch.save(model.state_dict(), "mnist_cnn.pt")
    # Save only on rank 0 to avoid rank 1 to overrides the checkpoint

    if args.use_ddp:
        # Teardown
        torch.distributed.destroy_process_group()
        # Teardown

    print(f"TIME SPENT: {time() - t0}")


if __name__ == '__main__':
    use_spawn = int(os.getenv("USE_SPAWN", 1))
    worldsize = int(os.getenv("WORLD_SIZE", 2))

    if use_spawn:
        # WORLD_SIZE=2 USE_SPAWN=1 python ddp_mnist_spawn/pytorch.py
        mp.spawn(main, args=(worldsize, use_spawn), nprocs=worldsize)
    else:
        # terminal 1: WORLD_SIZE=2 LOCAL_RANK=1 python ddp_mnist_spawn/pytorch.py
        # terminal 2: WORLD_SIZE=2 LOCAL_RANK=0 python ddp_mnist_spawn/pytorch.py
        main(int(os.getenv("LOCAL_RANK")), worldsize, use_spawn)