-
Notifications
You must be signed in to change notification settings - Fork 389
delete delayed scaling from torchao.float8 #1753
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
Conversation
|
Stack from ghstack (oldest at bottom): |
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1753
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit c95ecdf with merge base dc0134e ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Summary: Previously we added a deprecation warning for delayed scaling in v0.9.0. Now that the branch cut for v0.9.0 has passed, we are deleting delayed scaling from the codebase to simplify the code. Motivation: 1. delayed scaling adds a lot of complexity 2. we still don't have any customers for delayed scaling, since people prefer to optimize for higher accuracy from dynamic scaling If a need for delayed scaling arises in the future, we can bring it back to a prototype folder in a way that is decoupled from the dynamic scaling code. Test Plan: ``` ./test/float8/test_everything.sh ``` Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 8d8b801 ghstack-comment-id: 2674970350 Pull Request resolved: #1753
|
cc @lw , it's happening! |
drisspg
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
🔥
Summary:
Previously we added a deprecation warning for delayed scaling in v0.9.0.
Now that the branch cut for v0.9.0 has passed, we are deleting delayed
scaling from the codebase to simplify the code.
Motivation:
prefer to optimize for higher accuracy from dynamic scaling
If a need for delayed scaling arises in the future, we can bring it back
to a prototype folder in a way that is decoupled from the dynamic
scaling code.
For more context, see #1680
Python LOC impact:
(source: https://gist.github.com/vkuzo/2c845165e7bfe8933cf1bc24561cd613)
Test Plan:
Reviewers:
Subscribers:
Tasks:
Tags: