PyTorch RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

with torch.set_grad_enabled(True)：

最近在跑一个项目时，遇到了标题中提到的报错。
报错代码片段如下：

X = X.clone().detach().requires_grad_(True).cuda()
pred = model(X, timesteps=timesteps)
mask = pred.argmax(-1).eq(y) # stop attacking when misclassified
if