Gated Attention & Scaling Speedruns #2281
Draft
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
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
…calvin/gated_attention
Member
Author
|
For some reason, not having the separate gate projection seems to be much more expensive FLOP-wise here, even after obtaining the results on v5p32. Will revert back to that and redo 2x and 2.5x sweep. 
|
let XLA do its magic & not mess up dimension size alignment
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Outdated suggestions cannot be applied.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.
Description
Addresses #2109
Implements Gated Attention per https://github.com/qiuzh20/gated_attention and sweeps to find optimal LR scaling factor.
Basically ready for a while but rerunning the 1.2B track on v5p-32 to get good hardware FLOPs data points, and we are having trouble launching v5p-32 specifically on our clusters. Putting a draft PR up to let people know this is being worked on lest effort be duplicated (Will almost did).
Checklist
uv run python infra/pre-commit.py --all-filesto lint/format your code