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

TODO: #31

Open

5 of 8 tasks

surcyf123 opened this issue Sep 10, 2023 · 0 comments

Assignees

Owner

surcyf123 commented Sep 10, 2023 •

edited

Loading

- isolate normalization and raw scoring on endpoints.
- Iterate through more sample completions to find a good constant mean and variance for each reward model
- organize branches- move scraping files to new feature branch and then PR to ethan's branch
- setup clear pipeline for benchmarking in the future- low priority
- fix cuda and allow the input tensors to be in both in fp16 and fp32 depending on the reward model

09/10

benchmark each thread of the reward model and 2. see if we can combine parts into only 2 GPUs than run 2 per 4x
figure out latency problems, decide a region to host everything
for testing, setup non threaded reward endpoints

surcyf123 self-assigned this

surcyf123 mentioned this issue

memory optimization on reward endpoints #19

Closed

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