pytorch保存模型pth_PyTorch之保存加载模型

Bertha 。 2022-12-31 15:30 315阅读 0赞

前提

SAVING AND LOADING MODELS

当提到保存和加载模型时，有三个核心功能需要熟悉：

1.torch.save：将序列化的对象保存到disk。这个函数使用Python的pickle实用程序进行序列化。使用这个函数可以保存各种对象的模型、张量和字典。

2.torch.load：使用pickle unpickle工具将pickle的对象文件反序列化为内存。

3.torch.nn.Module.load\_state\_dict:使用反序列化状态字典加载model’s参数字典。

一：WHAT IS A STATE\_DICT

在PyTorch中，torch.nn.Module的可学习参数(即权重和偏差)，模块模型包含在model's参数中(通过model.parameters()访问)。state\_dict是个简单的Python dictionary对象，它将每个层映射到它的参数张量。

注意，只有具有可学习参数的层(卷积层、线性层等)才有model's state\_dict中的条目。优化器对象(connector .optim)也有一个state\_dict，其中包含关于优化器状态以及所使用的超参数的信息。

Example:

import torch

import torch.nn as nn

import torch.nn.functional as F

\# Define model

class TheModelClass(nn.Module):

def \_\_init\_\_(self):

super(TheModelClass,self).\_\_init\_\_()

self.conv1=nn.Conv2d(3,6,5)

self.pool=nn.MaxPool2d(2,2)

self.conv2=nn.Conv2d(6,16,5)

self.fc1=nn.Linear(16\*5\*5,120)

self.fc2=nn.Linear(120,84)

self.fc3=nn.Linear(84,10)

def farward(self,x):

x=self.pool(F.relu(self.conv1(x)))

x=self.pool(F.relu(self.conv2(x)))

x=x.view(-1,16\*5\*5)

x=F.relu(self.fc1(x))

x=F.relu(self.fc2(x))

x=self.fc3(x)

return x

\# Initialize model

model=TheModelClass()

\# Initialize optimizer

optimizer=torch.optim.SGD(model.parameters(),lr=1e-4,momentum=0.9)

print("Model's state\_dict:")

\# Print model's state\_dict

for param\_tensor in model.state\_dict():

print(param\_tensor,"\\t",model.state\_dict()\[param\_tensor\].size())

print("optimizer's state\_dict:")

\# Print optimizer's state\_dict

for var\_name in optimizer.state\_dict():

print(var\_name,"\\t",optimizer.state\_dict()\[var\_name\])

Output:

Model's state\_dict:

conv1.weight torch.Size(\[6, 3, 5, 5\])

conv1.bias torch.Size(\[6\])

conv2.weight torch.Size(\[16, 6, 5, 5\])

conv2.bias torch.Size(\[16\])

fc1.weight torch.Size(\[120, 400\])

fc1.bias torch.Size(\[120\])

fc2.weight torch.Size(\[84, 120\])

fc2.bias torch.Size(\[84\])

fc3.weight torch.Size(\[10, 84\])

fc3.bias torch.Size(\[10\])

optimizer's state\_dict:

state \{\}

param\_groups \[\{'lr': 0.0001, 'momentum': 0.9, 'dampening': 0, 'weight\_decay': 0, 'nesterov': False, 'params': \[1310469552240, 1310469552384, 1310469552456, 1310469552528, 1310469552600, 1310469552672, 1310469552744, 1310469552816, 1310469552888, 1310469552960\]\}\]

二：SAVING & LOADING MODEL FOR INFERENCE

Save/Load state\_dict (Recommended)

Save:

torch.save(model.state\_dict(), PATH)

在保存模型进行推理时，只需要保存训练过的模型的学习参数即可。一个常见的PyTorch约定是使用.pt或.pth文件扩展名保存模型。

Load:

model = TheModelClass(\*args, \*\*kwargs)

model.load\_state\_dict(torch.load(PATH))

model.eval()

记住，您必须调用model.eval()，以便在运行推断之前将dropout和batch规范化层设置为评估模式。如果不这样做，将会产生不一致的推断结果。

Note:

注意，load\_state\_dict()函数接受一个dictionary对象，而不是保存对象的路径。这意味着您必须在将保存的state\_dict传至load\_state\_dict()函数之前反序列化它。

Save/Load Entire Model

Save:

torch.save(model, PATH)

Load:

\# Model class must be defined somewhere

model = torch.load(PATH)

model.eval()

三：

Save:

torch.save(\{

'epoch': epoch,

'model\_state\_dict': model.state\_dict(),

'optimizer\_state\_dict': optimizer.state\_dict(),

'loss': loss,

...

\}, PATH)

Load:

model = TheModelClass(\*args, \*\*kwargs)

optimizer = TheOptimizerClass(\*args, \*\*kwargs)

checkpoint = torch.load(PATH)

model.load\_state\_dict(checkpoint\['model\_state\_dict'\])

optimizer.load\_state\_dict(checkpoint\['optimizer\_state\_dict'\])

epoch = checkpoint\['epoch'\]

loss = checkpoint\['loss'\]

model.eval()

\# - or -

model.train()

在保存用于推理或恢复训练的通用检查点时，必须保存模型的state\_dict。另外，保存优化器的state\_dict也是很重要的，因为它包含缓冲区和参数，这些缓冲区和参数是在模型训练时更新的。要保存多个组件，请将它们组织在字典中，并使用torch.save()序列化字典。一个常见的PyTorch约定是使用.tar文件扩展名保存这些检查点。

四：SAVING & LOADING MODEL ACROSS DEVICES

Save on GPU, Load on CPU

Save:

torch.save(model.state\_dict(), PATH)

Load:

device = torch.device('cpu')

model = TheModelClass(\*args, \*\*kwargs)

model.load\_state\_dict(torch.load(PATH, map\_location=device))

Save on GPU, Load on GPU

Save:

torch.save(model.state\_dict(), PATH)

Load: