MXNet Error when using d2l.train_and_predict_rnn_gluon

I have encountered error when using d2l.train_and_predict_rnn_gluon, it produces the following message. Any hint on how to resolve it? Some resource points out that it is because multiple backward() are performed. But I don’t think the implementation of this function involves multiple backward(). I’m using MXNet 1.2.1 and d2l 0.8.7.

Can you send me the code, so I can regenerate the error?

Thank you Rachel! Update mxnet to 1.3.1 solves this problem