pytorch 多卡训练--DataParallel
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
- 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
- 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。
- 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：
import torch
from torchvision import datasets, transforms
import torchvision
from tqdm import tqdm
device_ids = [0, 1, 2, 3, ] # 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。
- 模型保存
使用nn.DataParallel包装后的模型参数的关键字会比没用nn.DataParallel包装的模型参数的关键字前面多一个“module.”，可以看看前面的错误，missing key 和 unexpected key的差别就在那个“module.”上。还有一种情况就是missing key 是module.features.0.weight，但unexpected key是 features.module.0.weight，就是module和feature的位置反过来了。
def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch):
checkpoint_path = join(
checkpoint_dir, "checkpoint_step{:09d}.pth".format(global_step))
optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None
torch.save({
"state_dict": model.model.state_dict(),#DataParallel保存模型会在key前面增加model，去掉方便单卡模型导入
"optimizer": optimizer_state,
"global_step": step,
"global_epoch": epoch,
}, checkpoint_path)
print("Saved checkpoint:", checkpoint_path)
def _load(checkpoint_path):
if use_cuda:
checkpoint = torch.load(checkpoint_path)
else:
checkpoint = torch.load(checkpoint_path,
map_location=lambda storage, loc: storage)
return checkpoint
def load_checkpoint(path, model, optimizer, reset_optimizer=False):
global global_step
global global_epoch
print("Load checkpoint from: {}".format(path))
checkpoint = _load(path)
from collections import OrderedDict
new_state_dict = OrderedDict()
state_dict =checkpoint["state_dict"] #手动模型参数关键字添加“module.”
for k, v in state_dict.items():
k = 'module.'+k
new_state_dict[k]=v
model.load_state_dict(new_state_dict)
if not reset_optimizer:
optimizer_state = checkpoint["optimizer"]
if optimizer_state is not None:
print("Load optimizer state from {}".format(path))
optimizer.load_state_dict(checkpoint["optimizer"])
global_step = checkpoint["global_step"]
global_epoch = checkpoint["global_epoch"]
return model