Error when i tried to train rnn

when i tried to train rnn with the command
python main.py --network_type rnn --dataset ptb --controller_optim adam --controller_lr 0.00035 \
--shared_optim sgd --shared_lr 20.0 --entropy_coeff 0.0001
one error occures at util.detach(h), it seems that when h is a tensor, this function will be infinitely recursive, causing an error.
Can you tell me why this is happening?

Reproduce the report using the command in main.py, then inspect util.detach(h) and how tensor h is passed during RNN training. Done means identifying the recursion cause and documenting or correcting the failing behavior; no specific test is named in the issue.

Written by the indexing model from the issue text.