-
Notifications
You must be signed in to change notification settings - Fork 423
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.
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
Speed monitor refactor #1987
Speed monitor refactor #1987
Conversation
Discussing offline with Vitaliy about if we want to include the cost estimate stuff. PR is being held until that is finalized. Would like review from examples repo folks that the changes ive made to speed monitor are fine in description |
Per discussion with @abhi-mosaic and @vchiley, we will not include the cost estimates. So this is ready for review! |
* add speed monitor refactor * fix docs * fix tests * fix remove 1 * extend test * format * respond to comments * restore caching * add deepcopy * add comment
* add speed monitor refactor * fix docs * fix tests * fix remove 1 * extend test * format * respond to comments * restore caching * add deepcopy * add comment
* add speed monitor refactor * fix docs * fix tests * fix remove 1 * extend test * format * respond to comments * restore caching * add deepcopy * add comment
What does this PR do?
Revamps Speed Monitor to track a whole bunch of metrics from per device to MFU. This is mostly upstreamed from examples repo
Changes:
flops_per_batch
, letting user specify assumptions of forward vs. backwardstate.dataloader.dataset.max_seq_len
Screenshot:
w diffusion models
What issue(s) does this change relate to?
CO-1761