Repository navigation
One Buffer per Attention unit - #40
Open
almogtavor wants to merge 1 commit into
Open
almogtavor wants to merge 1 commit into
almogtavor wants to merge 1 commit into
Conversation
Each Attention unit gets its own Gloo subgroup and Buffer, so an exchange or a close on one unit involves only the FFN ranks and that unit. The FFN ranks still wait for every unit they receive from. Lanes and Units share the helpers that give each Buffer its own communication stream.
|
This branch has not been deployed
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Step 1 of #38.
Goal: each Attention unit writes to its own Buffer, so a unit can be added or removed without rebuilding the Buffers of the others.
Units(group, config, units)builds one Buffer per Attention unit over the same FFN ranks. Nothing else changes: the FFN ranks still wait for every unit.Results (8x H100, one NVLink domain, exact match with the single-Buffer reference, eager and CUDA graph):
In every run the last unit closes its Buffer and the other units keep exchanging on theirs.
The existing suite (
tests.gpu.deepep --lanes 2, BF16 and FP8) also passes, and so do the 59 unit tests and Ruff.Not included: DBO inside a unit, and adding a unit while the others run.