PyTorch翻译官网教程8-SAVE AND LOAD THE MODEL-CSDN博客

本文链接：https://blog.csdn.net/poorCoder_/article/details/131737315

本文介绍了如何在PyTorch中保存和加载模型的权重，通过`state_dict`方法持久化模型学习到的参数，并展示了加载模型时需先实例化模型并调用`load_state_dict()`。此外，还提及了保存整个模型结构与权重的方法，以及加载时对模型状态的调整，如设置为评估模式。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

官网链接

Save and Load the Model — PyTorch Tutorials 2.0.1+cu117 documentation

保存和加载模型

在本节中，我们将了解如何通过保存、加载和运行模型预测来持久化模型状态。

import torch
import torchvision.models as models

保存和加载模型权重

PyTorch模型将学习到的参数存储在一个名为state_dict的内部状态字典中。这些可以通过torch.save 方法持久化

model = models.vgg16(weights='IMAGENET1K_V1')
torch.save(model.state_dict(), 'model_weights.pth')

输出

Downloading: "https://download.pytorch.org/models/vgg16-397923af.pth" to /var/lib/jenkins/.cache/torch/hub/checkpoints/vgg16-397923af.pth

  0%|          | 0.00/528M [00:00<?, ?B/s]
  5%|4         | 23.9M/528M [00:00<00:02, 250MB/s]
 10%|9         | 50.7M/528M [00:00<00:01, 268MB/s]
 16%|#6        | 85.5M/528M [00:00<00:01, 313MB/s]
 22%|##2       | 118M/528M [00:00<00:01, 322MB/s]
 28%|##8       | 148M/528M [00:00<00:01, 304MB/s]
 34%|###3      | 178M/528M [00:00<00:01, 292MB/s]
 39%|###8      | 206M/528M [00:00<00:01, 285MB/s]
 44%|####4     | 233M/528M [00:00<00:01, 267MB/s]
 49%|####8     | 258M/528M [00:00<00:01, 267MB/s]
 54%|#####3    | 284M/528M [00:01<00:00, 269MB/s]
 59%|#####8    | 310M/528M [00:01<00:00, 269MB/s]
 64%|######3   | 336M/528M [00:01<00:00, 270MB/s]
 69%|######8   | 362M/528M [00:01<00:00, 269MB/s]
 74%|#######3  | 388M/528M [00:01<00:00, 270MB/s]
 78%|#######8  | 414M/528M [00:01<00:00, 270MB/s]
 83%|########3 | 440M/528M [00:01<00:00, 270MB/s]
 88%|########8 | 465M/528M [00:01<00:00, 270MB/s]
 93%|#########3| 491M/528M [00:01<00:00, 270MB/s]
 98%|#########8| 517M/528M [00:01<00:00, 271MB/s]
100%|##########| 528M/528M [00:02<00:00, 276MB/s]

要加载模型权重，需要首先创建同一模型的实例，然后使用load_state_dict() 方法加载参数。

model = models.vgg16() # we do not specify ``weights``, i.e. create untrained model
model.load_state_dict(torch.load('model_weights.pth'))
model.eval()

输出

VGG(
  (features): Sequential(
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (18): ReLU(inplace=True)
    (19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (25): ReLU(inplace=True)
    (26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (27): ReLU(inplace=True)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    (6): Linear(in_features=4096, out_features=1000, bias=True)
  )
)

注意

一定要在推理之前调用model.eval() 方法，将dropout层和normalization 层设置成evaluation模式。如果不这样做，将产生不一致的推理结果。