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

我是在修改源码后发现这个问题，排除各种原因后，发现是输入数据的问题

for epoch in range(epochs, n_epochs):
    for i, batch in enumerate(loader):

        # Model inputs
        real_A = batch[0].cuda()
        real_B = batch[1].cuda()
        real_P = batch[2].cuda()

RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn
然后修改为下面，就成功运行了

for epoch in range(epochs, n_epochs):
    for i, batch in enumerate(loader):

        # Model inputs
        real_A = batch[0].cuda().requires_grad_(True)
        real_B = batch[1].cuda().requires_grad_(True)
        real_P = batch[2].cuda().requires_grad_(True)