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

Incorrect taskset benchmark lower bounds #22

Open

theodorju opened this issue Mar 14, 2025 · 0 comments

theodorju commented Mar 14, 2025

Hi,

I think the following lower bounds here are incorrectly set for taskset (based on the paper and implementation)

learning rate lower bound should be 1e-8 instead of 1e-9
beta2 lower bound should be 1e-6 instead of 1e-3
linear decay lower bound should be 1e-7 instead of 1e-8
l1 and l2 lower bounds should be 1e-8 instead of 1e-9
epsilon lower bound should be 1e-10 instead of 1e-12

These boundaries are mentioned in the TaskSet paper on section "D.1. Adam8p, Adam6p, Adam4p, AdamLr search spaces".

They are also committed like this on the taskset repo.

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