%matplotlib inline
数据并行
Pytorch可以使用多个GPU,如把一个模型放置到GPU上
device = torch.device('cuda:0')
model.to(device)
GPU:然后复制所有的张量到GPU上:
mtensor = my_tensor.to(device)
只调用my_tensor.to(device)并没有复制张量到GPU上,而是返回一个copy。所以需要赋值一个新的张量并在GPU上使用该张量。
Pytorch默认只使用一个GPU,使用DataParallel可以让模型在多个GPU上运行
import torch
import torch.nn as nn
from torch.utils.data import DataLoader, Dataset
input_size = 5
output_size = 2
batch_size = 30
data_size = 100
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
class RandomDataset(Dataset):
def __init__(self, size, length):
self.len = length
self.data = torch.randn(length, size)
def __getitem__(self, index):
return self.data[index]
def __len__(self):
return self.len
rand_loader = DataLoader(dataset=RandomDataset(input_size, data_size),
batch_size=batch_size,
shuffle=True)
class Net(nn.Module):
def __init__(self, input_size, output_size):
super(Net, self).__init__()
self.fc = nn.Linear(input_size, output_size)
def forward(self, input):
output = self.fc(input)
print('\tIn Model: input size:', input_size, 'output_size:', output_size)
return output
创建实例并检测是由有多个GPU,如有多个则使用nn.DataParallel把模型放到GPU上
model = Net(input_size, output_size)
if torch.cuda.device_count() > 1:
print('Let us use:', torch.cuda.device_count(), 'GPUs!')
model = nn.DataParallel(model)
model.to(device)
Net(
(fc): Linear(in_features=5, out_features=2, bias=True)
)
for data in rand_loader:
input = data.to(device)
output = model(input)
print('outside: input_size', input.size(), 'output_size:', output.size())
In Model: input size: 5 output_size: 2
outside: input_size torch.Size([30, 5]) output_size: torch.Size([30, 2])
In Model: input size: 5 output_size: 2
outside: input_size torch.Size([30, 5]) output_size: torch.Size([30, 2])
In Model: input size: 5 output_size: 2
outside: input_size torch.Size([30, 5]) output_size: torch.Size([30, 2])
In Model: input size: 5 output_size: 2
outside: input_size torch.Size([10, 5]) output_size: torch.Size([10, 2])