Cudnn_status_bad_param

I am stuck in LSTM with CUDA, which is the same as cudnn-lstm.
I don’t know if this problem has already been solved since there is no solution yet.
Any suggestions?