训练报错：RuntimeError: CUDA error: device-side assert triggered

训练分类网络resnet18时发生报错：RuntimeError: CUDA error: device-side assert triggered

原始代码：

for batch_index, (images, labels) in enumerate(train_dataloader):
    if epoch <= args.warm:
        warmup_scheduler.step()
    images = images.cuda()
    labels = labels.cuda()

    optimizer.zero_grad()
    predicts = net(images)

    loss = 0
    for i, pi in enumerate(predicts):
        t = torch.full_like(pi, 0, device=device)
                
        if labels[i].long() > 0:
            print('labels:', labels[i])
            t[labels[i].long() - 1] = 1
            print('t', t)
        loss_curr = criterion(pi, t) # BCE
        loss += loss_curr
        train_total_loss += loss_curr.item()

        loss.backward()
        optimizer.step()

 发生报错后，将报错位置进行打印：

labels: tensor([7], device='cuda:0', dtype=torch.int32)
t tensor([0.