pytorch是一種比較吃顯存的python應(yīng)用。有些時(shí)候我們調(diào)試代碼需要關(guān)注顯存占用量,在調(diào)試代碼的時(shí)候也有查看網(wǎng)絡(luò)參數(shù)的需要。這些調(diào)試是需要的參數(shù)我們?cè)谡{(diào)試工具中就能看到,那么有哪些調(diào)試工具可以看這些參數(shù)呢?又該怎么看這些參數(shù)呢?接下來(lái)的這篇文章帶你了解。
1.使用torchstat
pip install torchstat
from torchstat import stat
import torchvision.models as models
model = models.resnet152()
stat(model, (3, 224, 224))
關(guān)于stat函數(shù)的參數(shù),第一個(gè)應(yīng)該是模型,第二個(gè)則是輸入尺寸,3為通道數(shù)。我沒(méi)有調(diào)研該函數(shù)的詳細(xì)參數(shù),也不知道為什么使用的時(shí)候并不提示相應(yīng)的參數(shù)。
2.使用torchsummary
pip install torchsummary
from torchsummary import summary
summary(model.cuda(),input_size=(3,32,32),batch_size=-1)
使用該函數(shù)直接對(duì)參數(shù)進(jìn)行提示,可以發(fā)現(xiàn)直接有顯式輸入batch_size的地方,我自己的感覺(jué)好像該函數(shù)更好一些。但是!?。〔恢罏槭裁?,該函數(shù)在我的機(jī)器上一直報(bào)錯(cuò)!??!
TypeError: can't convert CUDA tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
Update:經(jīng)過(guò)論壇咨詢,報(bào)錯(cuò)的原因找到了,只需要把
pip install torchsummary
修改為
pip install torch-summary
補(bǔ)充:Pytorch查看模型參數(shù)并計(jì)算模型參數(shù)量與可訓(xùn)練參數(shù)量
查看模型參數(shù)(以AlexNet為例)
import torch
import torch.nn as nn
import torchvision
class AlexNet(nn.Module):
def __init__(self,num_classes=1000):
super(AlexNet,self).__init__()
self.feature_extraction = nn.Sequential(
nn.Conv2d(in_channels=3,out_channels=96,kernel_size=11,stride=4,padding=2,bias=False),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=3,stride=2,padding=0),
nn.Conv2d(in_channels=96,out_channels=192,kernel_size=5,stride=1,padding=2,bias=False),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=3,stride=2,padding=0),
nn.Conv2d(in_channels=192,out_channels=384,kernel_size=3,stride=1,padding=1,bias=False),
nn.ReLU(inplace=True),
nn.Conv2d(in_channels=384,out_channels=256,kernel_size=3,stride=1,padding=1,bias=False),
nn.ReLU(inplace=True),
nn.Conv2d(in_channels=256,out_channels=256,kernel_size=3,stride=1,padding=1,bias=False),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=3, stride=2, padding=0),
)
self.classifier = nn.Sequential(
nn.Dropout(p=0.5),
nn.Linear(in_features=256*6*6,out_features=4096),
nn.ReLU(inplace=True),
nn.Dropout(p=0.5),
nn.Linear(in_features=4096, out_features=4096),
nn.ReLU(inplace=True),
nn.Linear(in_features=4096, out_features=num_classes),
)
def forward(self,x):
x = self.feature_extraction(x)
x = x.view(x.size(0),256*6*6)
x = self.classifier(x)
return x
if __name__ =='__main__':
# model = torchvision.models.AlexNet()
model = AlexNet()
# 打印模型參數(shù)
#for param in model.parameters():
#print(param)
#打印模型名稱與shape
for name,parameters in model.named_parameters():
print(name,':',parameters.size())
feature_extraction.0.weight : torch.Size([96, 3, 11, 11])
feature_extraction.3.weight : torch.Size([192, 96, 5, 5])
feature_extraction.6.weight : torch.Size([384, 192, 3, 3])
feature_extraction.8.weight : torch.Size([256, 384, 3, 3])
feature_extraction.10.weight : torch.Size([256, 256, 3, 3])
classifier.1.weight : torch.Size([4096, 9216])
classifier.1.bias : torch.Size([4096])
classifier.4.weight : torch.Size([4096, 4096])
classifier.4.bias : torch.Size([4096])
classifier.6.weight : torch.Size([1000, 4096])
classifier.6.bias : torch.Size([1000])
計(jì)算參數(shù)量與可訓(xùn)練參數(shù)量
def get_parameter_number(model):
total_num = sum(p.numel() for p in model.parameters())
trainable_num = sum(p.numel() for p in model.parameters() if p.requires_grad)
return {'Total': total_num, 'Trainable': trainable_num}
第三方工具
from torchstat import stat
import torchvision.models as models
model = models.alexnet()
stat(model, (3, 224, 224))
from torchvision.models import alexnet
import torch
from thop import profile
model = alexnet()
input = torch.randn(1, 3, 224, 224)
flops, params = profile(model, inputs=(input, ))
print(flops, params)
以上就是pytorch查看網(wǎng)絡(luò)參數(shù)顯存占用量等操作的全部?jī)?nèi)容,希望能給大家一個(gè)參考,也希望大家多多支持W3Cschool。如有錯(cuò)誤或未考慮完全的地方,望不吝賜教。