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

differentiable top k #7

Open

wangzizhao opened this issue Feb 20, 2024 · 0 comments

wangzizhao commented Feb 20, 2024 •

edited

Loading

IIUC, the topk in colt5_attention uses coor_descent, and, according to the original paper Eq 8 - 11, it seems to expect the input to be unnormalized.

However, in the forward of TopNGating, it seems that normalized score is passed into the topk.

I wonder if I misunderstood something and whether I should use normalized or unnormalized score here.

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