[WIP] fused sdpa training #4498
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
github-actions
bot
added
platform:gpu-intel
syurkevi
force-pushed
the
syurkevi/fused_sdpa_training
branch
from
2bcb1f5 to
adf7f0e
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.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
This is a WIP PR implementing a fused sdpa training kernel. The current bottleneck seems to be the individual gemm ukernels which with their current setup are ~3-5 slower than the primitives based gemms of the same size. Tuning allows for a large variation in runtime, however certain tile sizes seem to be inaccessible due to hardcoded FMA values in microkernel_provider.cpp
Ongoing work is focusing on improving gemmstone strategies to more closely match the performance of standalone gemms. Once those are closer the benefits of the fused kernel can be better realized.