Merged
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
Reverts prior Triton module/function rename by switching call sites back to the *_fwd.py / *_bwd.py module naming, and reintroducing dedicated forward-combine and backward pre/post-processing modules used by the attention kernels.
Changes:
- Update imports in tests and Triton interface to use
flash_*_{fwd,bwd}modules. - Replace previous combine/preprocess/postprocess module references with
flash_fwd_combine,flash_bwd_preprocess, andflash_bwd_postprocess. - Add new Triton kernels/modules for forward split-KV combine and backward preprocess/postprocess.
Reviewed changes
Copilot reviewed 8 out of 11 changed files in this pull request and generated no comments.
Show a summary per file
| File | Description |
|---|---|
| tests/test_utils.py | Updates test imports to the reverted *_fwd/*_bwd module names. |
| flash_sparse_attn/ops/triton/interface.py | Updates interface imports to the reverted module names. |
| flash_sparse_attn/ops/triton/flash_sparse_fwd.py | Switches split-KV combine call to flash_fwd_combine. |
| flash_sparse_attn/ops/triton/flash_sparse_bwd.py | Switches backward pre/post steps to flash_bwd_preprocess/flash_bwd_postprocess. |
| flash_sparse_attn/ops/triton/flash_gated_fwd.py | Switches split-KV combine call to flash_fwd_combine. |
| flash_sparse_attn/ops/triton/flash_gated_bwd.py | Switches backward pre/post steps to flash_bwd_preprocess/flash_bwd_postprocess. |
| flash_sparse_attn/ops/triton/flash_dense_fwd.py | Switches split-KV combine call to flash_fwd_combine. |
| flash_sparse_attn/ops/triton/flash_dense_bwd.py | Switches backward pre/post steps to flash_bwd_preprocess/flash_bwd_postprocess. |
| flash_sparse_attn/ops/triton/flash_fwd_combine.py | Adds Triton kernel + wrapper to combine split-KV partial outputs/LSE into final output/LSE. |
| flash_sparse_attn/ops/triton/flash_bwd_preprocess.py | Adds Triton kernel + wrapper to compute dpsum, lse_log2, and initialize dq_accum. |
| flash_sparse_attn/ops/triton/flash_bwd_postprocess.py | Adds Triton kernel + wrapper to scale/store dq (and optional da) from accumulators. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
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.
Reverts #268