PyTorch-GPU加速实例

傷城~ 2022-12-27 01:51 216阅读 0赞

更多编程教程请到：[菜鸟教程][Link 1] https://www.piaodoo.com/

友情链接：

[高州阳光论坛][Link 2]https://www.hnthzk.com/

[人人影视][Link 3]http://www.op-kg.com/

硬件：NVIDIA-GTX1080

软件：Windows7、python3.6.5、pytorch-gpu-0.4.1

**一、基础知识**

将数据和网络都推到GPU，接上.cuda()

**二、代码展示**

import torch
    import torch.nn as nn
    import torch.utils.data as Data
    import torchvision
    # torch.manual_seed(1)

EPOCH = 1  
BATCH\_SIZE = 50  
LR = 0.001  
DOWNLOAD\_MNIST = False

train\_data = torchvision.datasets.MNIST(root=’./mnist/’, train=True, transform=torchvision.transforms.ToTensor(), download=DOWNLOAD\_MNIST,)  
train\_loader = Data.DataLoader(dataset=train\_data, batch\_size=BATCH\_SIZE, shuffle=True)

test\_data = torchvision.datasets.MNIST(root=’./mnist/’, train=False)

# !!! Change in here !!! #

test\_x = torch.unsqueeze(test\_data.test\_data, dim=1).type(torch.FloatTensor)\[:2000\].cuda()/255. \# Tensor on GPU  
test\_y = test\_data.test\_labels\[:2000\].cuda()

class CNN(nn.Module):  
def **init**(self):  
super(CNN, self).**init**()  
self.conv1 = nn.Sequential(nn.Conv2d(in\_channels=1, out\_channels=16, kernel\_size=5, stride=1, padding=2,),  
nn.ReLU(), nn.MaxPool2d(kernel\_size=2),)  
self.conv2 = nn.Sequential(nn.Conv2d(16, 32, 5, 1, 2), nn.ReLU(), nn.MaxPool2d(2),)  
self.out = nn.Linear(32 \* 7 \* 7, 10)

def forward(self, x):  
x = self.conv1(x)  
x = self.conv2(x)  
x = x.view(x.size(0), -1)  
output = self.out(x)  
return output

cnn = CNN()

# !!! Change in here !!! #

cnn.cuda() \# Moves all model parameters and buffers to the GPU.

optimizer = torch.optim.Adam(cnn.parameters(), lr=LR)  
loss\_func = nn.CrossEntropyLoss()

for epoch in range(EPOCH):  
for step, (x, y) in enumerate(train\_loader):

# !!! Change in here !!! #

b\_x = x.cuda() \# Tensor on GPU  
b\_y = y.cuda() \# Tensor on GPU

output = cnn(b\_x)  
loss = loss\_func(output, b\_y)  
optimizer.zero\_grad()  
loss.backward()  
optimizer.step()

if step % 50 == 0:  
test\_output = cnn(test\_x)

# !!! Change in here !!! #

pred\_y = torch.max(test\_output, 1)\[1\].cuda().data \# move the computation in GPU

accuracy = torch.sum(pred\_y == test\_y).type(torch.FloatTensor) / test\_y.size(0)  
print('Epoch: ', epoch, ‘| train loss: %.4f’ % loss, ‘| test accuracy: %.2f’ % accuracy)

test\_output = cnn(test\_x\[:10\])

# !!! Change in here !!! #

pred\_y = torch.max(test\_output, 1)\[1\].cuda().data \# move the computation in GPU

print(pred\_y, ‘prediction number’)  
print(test\_y\[:10\], ‘real number’)

**三、结果展示**

![Image 1][]

**补充知识：****pytorch使用gpu对网络计算进行加速**

**1.基本要求**

你的电脑里面有合适的GPU显卡（NVIDA）,并且需要支持CUDA模块

你必须安装GPU版的Torch，（详细安装方法请移步pytorch官网）

**2.使用GPU训练CNN**

利用pytorch使用GPU进行加速方法主要就是将数据的形式变成GPU能读的形式，然后将CNN也变成GPU能读的形式，具体办法就是在后面加上.cuda()。

例如：

#如何检查自己电脑是否支持cuda
    print torch.cuda.is_available()
    # 返回True代表支持，False代表不支持
    '''
    注意在进行某种运算的时候使用.cuda()
    '''
    test_data=test_data.test_labels[:2000].cuda()
    '''
    对于CNN与损失函数利用cuda加速
    '''
    class CNN(nn.Module):
     ...
    cnn=CNN()
    cnn.cuda()
    loss_f = t.nn.CrossEntropyLoss()
    loss_f = loss_f.cuda()

而在train时，对于train\_data训练过程进行GPU加速。也同样+.cuda()。

for epoch ..:
     for step, ...:
     1
    '''
    若你的train_data在训练时需要进行操作
    若没有其他操作仅仅只利用cnn()则无需另加.cuda()
    '''
    #eg
     train_data = torch.max(teain_data, 1)[1].cuda()

**补充：**取出数据需要从GPU切换到CPU上进行操作

**eg：**

> loss = loss.cpu()  
> acc = acc.cpu()

理解并不全，如有纰漏或者错误还望各位大佬指点迷津

以上这篇PyTorch-GPU加速实例就是小编分享给大家的全部内容了，希望能给大家一个参考，也希望大家多多支持菜鸟教程www.piaodoo.com。

[Link 1]: https://www.piaodoo.com/
[Link 2]: https://www.hnthzk.com/
[Link 3]: http://www.op-kg.com/
[Image 1]: