method one:
Save and read the entire model:
torch.save(model, PATH)
model = torch.load(PATH)
Method Two:
Only save the model's state_dict(), read:
torch.save(model.state_dict(), PATH)
state_dict = torch.load(PATH)
At this time, only the state_dict is read, and you need
new_model.load_state_dict(torch.load(PATH)) # 其中的new_model是新初始化的模型
problem
Damn , directly saving the model overturned, and an error similar to this kind of error occurred. In the future, I will only save the state_dict.