Open likuanppd opened 1 year ago
Hi! I write an issue in #128, which seems to be helpful in avoiding OOM in MetaAttack. I didn't check the solution thoroughly on different devices. Please let me know if it could be also helpful to you. Thanks.
I've tried your solution, and it works perfectly. Good job!
Hi jin,
we find that MetaAttack will face the OOM problem if the PyTorch version is higher than 1.9. Can you fix this problem? For example, we attack polblogs under 20% ptb rate in one tesla-v100 (32GB), and then the cuda will OOM. We retry this in another Linux server and it reports the same error, but if we turn the pytorch version into 1.8.2, everything would be ok.
Thanks