Pytorch多多GPU的计算和的计算和Sync BatchNorm
nn.DataParallel
pytorch中使用GPU非常方便和简单:
import torch
import torch.nn as nn
input_size = 5
output_size = 2
class Model(nn.Module):
def __init__(self, input_size, output_size):
super(Model, self).__init__()
self.fc = nn.Linear(input_size, output_size)
def forward(self, input):
output = self.fc(input)
print("[In Model]: device",torch.cuda.current_device() ," input size", input.size()," output size", output.size())
return output
device = torch.device('cuda:0')
model = Model(input_size, output_size)
model.to(device)
x = torch.Tensor(2,5)
x = x.to(device)
y = model(x)
这里需要注意的是,仅仅调用Tensor.to()只会在GPU上返回一个新的copy,并不会对原来的引用造成变化,因此需要通过赋值
rewrite。
上述只是对单个GPU的使用方法,对于多个GPU,pytorch也提供了封装好的接口——DataParallel,只需要将model 对象放入容器
中即可:
model = Model(input_size, output_size)
print("Let's use", torch.cuda.device_count(), "GPUs!")
model = nn.DataParallel(model)
model.to(device)
print(model)
# output
Let's use 2 GPUs!
DataParallel(
(module): Model(
(fc): Linear(in_features=5, out_features=2, bias=True)
)
)
看到这次输出的model外面还有一层DataParallel,但这里并没有体现出存在多个GPU。
接下来构造一个Dummy DataSet,来跑一下模型:
from torch.utils.data import Dataset, DataLoader
batch_size = 30
data_size = 100
class RandomDataset(Dataset):
def __init__(self, size, length):
self.len = length
self.data = torch.randn(length, size) # 有length个样本,每个样本是size长度的向量
评论0