pytorch多卡训练(含demo)

本文写于三年前，由于pytorch升级迭代，目前更推荐采用DDP实现方法，可看我新整理的一篇文章《Pytorch实现多GPU并行训练（DDP）_pytorch 多gpu并行训练-CSDN博客》

---

用pytorch进行多GPU训练，只需要学会把单卡训练的代码稍微改一下即可。不用弄得太麻烦。通过一个demo来做是最快入手的。

1. 要知道机器有几张卡： 

nvidia-smi

 2. 模型用DataParallel包装一下：

device_ids = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9] # 10卡机
model = torch.nn.DataParallel(model, device_ids=device_ids) # 指定要用到的设备
model = model.cuda(device=device_ids[0]) # 模型加载到设备0

3. 数据也指定设备：

X_train, y_train = X_train.cuda(device=device_ids[0]), y_train.cuda(device=device_ids[0])

这里只需要用device_ids[0]定义一个样式就好，不需要逐卡指定设备。但没这一步会报错。 

4. 最后，来看一个完整demo，有注释的地方就是与单卡训练不一样的地方：

（代码来自https://www.cnblogs.com/fanghao/p/10334631.html，原代码在我的环境中运行有bug，下面的程序是修改版）

import torch
from torchvision import datasets, transforms
import torchvision
from tqdm import tqdm

device_ids = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9] # 可用GPU
BATCH_SIZE = 64

transform = transforms.Compose([transforms.ToTensor()])
data_train = datasets.MNIST(root = "./data/",
                            transform=transform,
                            train=True,
                            download=True)
data_test = datasets.MNIST(root="./data/",
                           transform=transform,
                           train=False)

data_loader_train = torch.utils.data.DataLoader(dataset=data_train,
                                                # 单卡batch size * 卡数
                                                batch_size=BATCH_SIZE * len(device_ids),
                                                shuffle=True,
                                                num_workers=2)

data_loader_test = torch.utils.data.DataLoader(dataset=data_test,
                                               batch_size=BATCH_SIZE * len(device_ids),
                                               shuffle=True,
                                               num_workers=2)


class Model(torch.nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = torch.nn.Sequential(
        torch.nn.Conv2d(1, 64, kernel_size=3, stride=1, padding=1),
        torch.nn.ReLU(),
        torch.nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1),
        torch.nn.ReLU(),
        torch.nn.MaxPool2d(stride=2, kernel_size=2),
    )
        self.dense = torch.nn.Sequential(
            torch.nn.Linear(14 * 14 * 128, 1024),
            torch.nn.ReLU(),
            torch.nn.Dropout(p=0.5),
            torch.nn.Linear(1024, 10)
    )
    def forward(self, x):
        x = self.conv1(x)
        x = x.view(-1, 14 * 14 * 128)
        x = self.dense(x)
        return x


model = Model()
# 指定要用到的设备
model = torch.nn.DataParallel(model, device_ids=device_ids)
# 模型加载到设备0
model = model.cuda(device=device_ids[0])

cost = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters())
from time import sleep
n_epochs = 50
for epoch in range(n_epochs):
    running_loss = 0.0
    running_correct = 0
    print("Epoch {}/{}".format(epoch, n_epochs))
    print("-"*10)
    for data in tqdm(data_loader_train):
        X_train, y_train = data
        # 指定设备0
        X_train, y_train = X_train.cuda(device=device_ids[0]), y_train.cuda(device=device_ids[0])
        outputs = model(X_train)
        _,pred = torch.max(outputs.data, 1)
        optimizer.zero_grad()
        loss = cost(outputs, y_train)

        loss.backward()
        optimizer.step()
        running_loss += loss.data.item()
        running_correct += torch.sum(pred == y_train.data)
    testing_correct = 0
    for data in data_loader_test:
        X_test, y_test = data
        # 指定设备1
        X_test, y_test = X_test.cuda(device=device_ids[0]), y_test.cuda(device=device_ids[0])
        outputs = model(X_test)
        _, pred = torch.max(outputs.data, 1)
        testing_correct += torch.sum(pred == y_test.data)
    print("Loss is:{:.4f}, Train Accuracy is:{:.4f}%, Test Accuracy is:{:.4f}".format(torch.true_divide(running_loss, len(data_train)),
                                                                                      torch.true_divide(100*running_correct, len(data_train)),
                                                                                      torch.true_divide(100*testing_correct, len(data_test))))
torch.save(model.state_dict(), "model_parameter.pkl")

通过上面程序改造的方法，便可以把单卡训练改造成了多卡训练。当然，有些自定义的sampler，dataloader可能也要相应改之，具体问题具体处理。

我在另一个训练任务上，完全激活了10卡同时训练：10个2080ti，没有炫卡的意思(逃