PyTorch (6-3) Distributed Applications with PyTorch (一機多卡)

本文來自官方教學 https://pytorch.org/tutorials/intermediate/dist_tuto.html。 這篇範例可能比較舊，所以我改了一些地方。

1. Packages:

import os
import torch
import torch.distributed as dist
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim

from math import ceil
from random import Random
from torch.multiprocessing import Process
from torchvision import datasets, transforms

2. Partition:

將資料分成多個 chunks，之後會設定 chunk 數量等於顯卡數量。

設定 chunk 數量等於顯卡數量有兩個好處:

1. 每個 chunk 資料量相同
2. 每個顯卡依照自己的 GPU 編號取 chunk

class Partition(object):
    """ Dataset-like object, but only access a subset of it. """

    def __init__(self, data, index):
        self.data = data
        self.index = index

    def __len__(self):
        return len(self.index)

    def __getitem__(self, index):
        data_idx = self.index[index]
        return self.data[data_idx]


class DataPartitioner(object):
    """ Partitions a dataset into different chuncks. """

    def __init__(self, data, sizes, seed=1234):
        self.data = data
        self.partitions = []
        rng = Random()
        rng.seed(seed)
        data_len = len(data)
        indexes = [x for x in range(0, data_len)]
        rng.shuffle(indexes)

        for frac in sizes:
            part_len = int(frac * data_len)
            self.partitions.append(indexes[0:part_len])
            indexes = indexes[part_len:]

    def use(self, partition):
        return Partition(self.data, self.partitions[partition])

將資料平均分給各個 ranks，128 是一個 batch constant。我們會把這個 batch 分給多個 rank，但基數就是 128。

def partition_dataset():
    """ Partitioning MNIST """
    dataset = datasets.MNIST(
        '../../data/MNIST',
        train=True,
        download=True,
        transform=transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize((0.1307, ), (0.3081, ))
        ]))
    size = dist.get_world_size()
    bsz = int(128 / float(size))
    partition_sizes = [1.0 / size for _ in range(size)]
    partition = DataPartitioner(dataset, partition_sizes)
    partition = partition.use(dist.get_rank())
    train_set = torch.utils.data.DataLoader(
        partition, batch_size=bsz, shuffle=True)
    return train_set, bsz

3. ConvNet:

簡單 ConvNet。

class Net(nn.Module):
    """ Network architecture. """

    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
        self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
        self.conv2_drop = nn.Dropout2d()
        self.fc1 = nn.Linear(320, 50)
        self.fc2 = nn.Linear(50, 10)

    def forward(self, x):
        x = F.relu(F.max_pool2d(self.conv1(x), 2))
        x = F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
        x = x.view(-1, 320)
        x = F.relu(self.fc1(x))
        x = F.dropout(x, training=self.training)
        x = self.fc2(x)
        return x

4. Distributed Application:

接下來各 gpu 會各自計算梯度再同時更新，過程如下。

我們通常會把 model 放在不同 gpu 上，但也可以放在同個 gpu 裡 (比較沒有實質的意義)。

def run(rank, size):
    """ Distributed Synchronous SGD Example """
    torch.manual_seed(1234)
    train_set, bsz = partition_dataset()

# [start-20180912-cooper-mod] #
    model = Net()
    model = model.cuda(rank)
    # model = model.cuda(0)
# [end-20180912-cooper-mod] #

# [start-20180912-cooper-add] #
    criterion = nn.CrossEntropyLoss()
# [end-20180912-cooper-add] #

    optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)

    num_batches = ceil(len(train_set.dataset) / float(bsz))
    for epoch in range(10):
        epoch_loss = 0.0
        for data, target in train_set:
# [start-20180912-cooper-mod] #
            data, target = torch.tensor(data), torch.tensor(target)
            data, target = data.cuda(rank), target.cuda(rank)
            # data, target = data.cuda(0), target.cuda(0)
# [end-20180912-cooper-mod] #
            optimizer.zero_grad()
            output = model(data)
# [start-20180912-cooper-mod] #
            loss = criterion(output, target)
            epoch_loss += loss.item()
            loss.backward()
# [end-20180912-cooper-mod] #
            average_gradients(model)
            optimizer.step()

        print("Rank: {:2}, epoch: {:2}, loss: {:.3}".format(
            dist.get_rank(), epoch, epoch_loss / num_batches))

加總並平均所有梯度後 optimizer 會更新所有 tensor。

def average_gradients(model):
    """ Gradient averaging. """
    size = float(dist.get_world_size())
    for param in model.parameters():
        dist.all_reduce(param.grad.data, op=dist.reduce_op.SUM, group=0)
        param.grad.data /= size

5. 初始化和運行：

根據官網的說法https://pytorch.org/docs/stable/distributed.html，backends 會決定傳輸的 tensor 類型。

選用 gloo:

def init_processes(rank, size, fn, backend='gloo'):
    """ Initialize the distributed environment. """
    os.environ['MASTER_ADDR'] = '127.0.0.1'
    os.environ['MASTER_PORT'] = '29500'
    dist.init_process_group(backend, rank=rank, world_size=size)
    fn(rank, size)

if __name__ == "__main__":
    size = 2
    processes = []
    for rank in range(size):
        p = Process(target=init_processes, args=(rank, size, run))
        p.start()
        processes.append(p)

    for p in processes:
        p.join()