New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

A question about "Loss is nan" #7

Closed

yafangna opened this issue Oct 11, 2022 · 1 comment

yafangna commented Oct 11, 2022

Is there any interface in the source code that would cause the loss is nan problem sometimes and the correct operation sometimes?

Owner

vtddggg commented Oct 12, 2022

"loss is nan" is a problem for original ViT models when amp is turned on. Check here for more details.

Afterwards, many techniques are proposed to solve this problem, e.g., LayerScale. You can refer these techniques for preventing training loss to nan.

vtddggg closed this as completed

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment