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

Broad Training: Design training loop #19

Open

bclavie opened this issue May 15, 2024 · 1 comment

Assignees

Labels

Contributor

bclavie commented May 15, 2024 •

edited

Loading

Broad issue that will eventually be split up. We need to experiment and implement what makes the most sense in terms of:

LR Schedule: Warmup-Stable-Decay or Reciprocal SQRT + Cosine to facilitate future continuous pre-training?
Context length ramp-up, to support our target context sizes (4k target for now)
Batch size ramp-up and figuring out what the best tradeoff is

bclavie added the Training label

ohallstrom self-assigned this

Collaborator

ohallstrom commented May 30, 2024

WSD has been added in the following PR: #31. I can have a go at the ramp-ups as well

ohallstrom mentioned this issue

Add custom trainer with support for batch rampup #76

Closed

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