Do FP8 rowwise bias addition in higher precision #4095
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D74408348 |
Summary: X-link: facebookresearch/FBGEMM#1179 Previously, when bias was used in our FP8 rowwise kernel, it was added to the accumulator in its native precision. For example, if the bias is bf16, we would do a bf16 + bf16 addition. However, it's a bit more efficient and a bit more accurate to leave the accumulator in fp32, cast the bias to fp32, then to an fp32 addition. Reviewed By: jianyuh Differential Revision: D74408348
|
This pull request was exported from Phabricator. Differential Revision: D74408348 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary: Previously, when bias was used in our FP8 rowwise kernel, it was added to the accumulator in its native precision. For example, if the bias is bf16, we would do a bf16 + bf16 addition. However, it's a bit more efficient and a bit more accurate to leave the accumulator in fp32, cast the bias to fp32, then to an fp32 addition.
Differential Revision: D74408348