Pytorch | Tutorial-07 儲存和載入模型

一碗给力嗯發表於2024-03-20
PyTorch模型

這是對 Pytorch 官網的 Tutorial 教程的中文翻譯。

在本節中，我們將瞭解如何透過儲存、載入和執行模型預測來持久儲存模型狀態。

import torch
import torchvision.models as models

儲存和載入模型權重

PyTorch 模型將學習到的引數儲存在內部狀態字典中，稱為 state_dict 。這些可以透過 torch.save 方法儲存：

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

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]
  2%|2         | 12.7M/528M [00:00<00:04, 133MB/s]
  5%|5         | 26.8M/528M [00:00<00:03, 142MB/s]
  8%|7         | 40.9M/528M [00:00<00:03, 145MB/s]
 10%|#         | 55.0M/528M [00:00<00:03, 146MB/s]
 13%|#3        | 69.1M/528M [00:00<00:03, 147MB/s]
 16%|#5        | 83.3M/528M [00:00<00:03, 147MB/s]
 18%|#8        | 97.4M/528M [00:00<00:03, 148MB/s]
 21%|##1       | 112M/528M [00:00<00:02, 148MB/s]
 24%|##3       | 126M/528M [00:00<00:02, 148MB/s]
 26%|##6       | 140M/528M [00:01<00:02, 148MB/s]
 29%|##9       | 154M/528M [00:01<00:02, 148MB/s]
 32%|###1      | 168M/528M [00:01<00:02, 148MB/s]
 35%|###4      | 182M/528M [00:01<00:02, 148MB/s]
 37%|###7      | 196M/528M [00:01<00:02, 148MB/s]
 40%|###9      | 211M/528M [00:01<00:02, 148MB/s]
 43%|####2     | 225M/528M [00:01<00:02, 148MB/s]
 45%|####5     | 239M/528M [00:01<00:02, 148MB/s]
 48%|####7     | 253M/528M [00:01<00:01, 148MB/s]
 51%|#####     | 267M/528M [00:01<00:01, 148MB/s]
 53%|#####3    | 281M/528M [00:02<00:01, 148MB/s]
 56%|#####5    | 295M/528M [00:02<00:01, 148MB/s]
 59%|#####8    | 309M/528M [00:02<00:01, 148MB/s]
 61%|######1   | 324M/528M [00:02<00:01, 148MB/s]
 64%|######3   | 338M/528M [00:02<00:01, 148MB/s]
 67%|######6   | 352M/528M [00:02<00:01, 148MB/s]
 69%|######9   | 366M/528M [00:02<00:01, 148MB/s]
 72%|#######2  | 380M/528M [00:02<00:01, 148MB/s]
 75%|#######4  | 394M/528M [00:02<00:00, 148MB/s]
 77%|#######7  | 408M/528M [00:02<00:00, 148MB/s]
 80%|########  | 423M/528M [00:03<00:00, 148MB/s]
 83%|########2 | 437M/528M [00:03<00:00, 148MB/s]
 85%|########5 | 451M/528M [00:03<00:00, 148MB/s]
 88%|########8 | 465M/528M [00:03<00:00, 148MB/s]
 91%|######### | 479M/528M [00:03<00:00, 142MB/s]
 93%|#########3| 493M/528M [00:03<00:00, 143MB/s]
 96%|#########6| 507M/528M [00:03<00:00, 144MB/s]
 99%|#########8| 521M/528M [00:03<00:00, 146MB/s]
100%|##########| 528M/528M [00:03<00:00, 147MB/s]

要載入模型權重，您需要先建立同一模型的例項，然後使用 load_state_dict() 方法載入引數。

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

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)
  )
)

請務必在推理之前呼叫 model.eval() 方法，將 dropout 和批次歸一化層設定為評估模式。如果不這樣做將會產生不一致的推理結果。

儲存和載入帶有形狀的模型

載入模型權重時，我們需要首先例項化模型類，因為該類定義了網路的結構。我們可能希望將此類的結構與模型一起儲存，在這種情況下，我們可以將 model （而不是 model.state_dict() ）傳遞給儲存函式：

torch.save(model, 'model.pth')

然後我們可以像這樣載入模型：

model = torch.load('model.pth')

此方法在序列化模型時使用 Python pickle 模組，因此它依賴於載入模型時實際可用的類的定義。

相關文章