CINN
CINN copied to clipboard
Add the implementation of relu_train to save mask to optimize the grad.
Thanks for your contribution!