@[toc]

问题描述

在重新加入模型，再训练的时候出现了如下问题：

Traceback (most recent call last):
  File "D:\Ghost_Demo\train.py", line 200, in <module>
    train_loss, train_acc = train(model_ft, DEVICE, train_loader, optimizer, epoch,model_ema)
  File "D:\Ghost_Demo\train.py", line 40, in train
    scaler.scale(loss).backward()
  File "D:\Users\wh109\anaconda3\lib\site-packages\torch\_tensor.py", line 396, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
  File "D:\Users\wh109\anaconda3\lib\site-packages\torch\autograd\__init__.py", line 173, in backward
    Variable._execution_engine.run_backward(  # Calls into the C++ engine to run the backward pass
RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

原因

出现这种错误是因为，构建Variable, 要注意得传入一个参数requires_grad=True, 这个参数表示是否对这个变量求梯度， 默认的是False, 也就是不对这个变量求梯度。

解决方法

将

data, target = data.to(device, non_blocking=True), target.to(device, non_blocking=True)

改为：

from torch.autograd import Variable
data, target = Variable(data, requires_grad=True).to(device, non_blocking=True), target.to(device,non_blocking=True)