当前位置：首页 > article >正文

PyTorch 保存和加载模型状态和优化器状态

article 2024/10/26 16:26:29

以下示例代码展示了如何在 PyTorch 中保存和加载模型状态和优化器状态，以便训练中断后可以继续训练。

1. 保存模型和优化器状态

假设模型训练了一段时间后，我们想要保存模型和优化器的状态，确保下次可以从同一位置继续训练。

2. 加载模型和优化器状态

加载保存的状态后，可以从保存的 epoch 继续训练。

示例代码

import torch
import torch.nn as nn
import torch.optim as optim

# 假设我们定义了一个简单的模型
class SimpleModel(nn.Module):
    def __init__(self):
        super(SimpleModel, self).__init__()
        self.fc = nn.Linear(10, 2)

    def forward(self, x):
        return self.fc(x)

# 创建模型和优化器
model = SimpleModel()
optimizer = optim.Adam(model.parameters(), lr=0.001)

# 模拟的训练代码片段
num_epochs = 20
checkpoint_path = "model_checkpoint.pth"

# 保存模型和优化器状态
def save_checkpoint(epoch, model, optimizer, path):
    torch.save({
        'epoch': epoch,
        'model_state_dict': model.state_dict(),
        'optimizer_state_dict': optimizer.state_dict()
    }, path)
    print(f"Checkpoint saved at epoch {epoch}.")

# 加载模型和优化器状态
def load_checkpoint(model, optimizer, path):
    checkpoint = torch.load(path)
    model.load_state_dict(checkpoint['model_state_dict'])
    optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
    start_epoch = checkpoint['epoch']
    print(f"Checkpoint loaded, starting at epoch {start_epoch}.")
    return start_epoch

# 尝试加载已保存的检查点
try:
    start_epoch = load_checkpoint(model, optimizer, checkpoint_path)
except FileNotFoundError:
    start_epoch = 0
    print("No checkpoint found, starting training from scratch.")

# 继续训练
for epoch in range(start_epoch, num_epochs):
    # 模拟训练步骤
    # output = model(input) ...
    # loss = loss_fn(output, target) ...
    # optimizer.zero_grad()
    # loss.backward()
    # optimizer.step()

    print(f"Epoch {epoch+1}/{num_epochs} completed.")

    # 每 5 个 epoch 保存一次模型状态
    if (epoch + 1) % 5 == 0:
        save_checkpoint(epoch + 1, model, optimizer, checkpoint_path)