Rename flash_dma to flash_dmattn and improve usability #61
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
Updates import statements and function references to use the more descriptive flash_dmattn naming convention instead of flash_dma. Ensures consistency across CUDA and Triton implementations while maintaining the same functionality and API structure.
Changes the description from "flash_attention" to "flash_dmattn" to reflect the correct kernel type being generated. Adds a default value of "instantiations" for the output directory parameter to improve usability when no specific directory is provided.
Updates package name throughout the build configuration to reflect the new flash_dmattn naming convention. Changes include package name, environment variable prefixes, CUDA extension name, and build exclusions. Reorganizes source file paths by moving instantiation files into a dedicated instantiations subdirectory for better code organization.
LoserCheems
requested review from
Evanwu1125,
SNHuan,
Copilot and
wubingheng111
and removed request for
Copilot
Contributor
There was a problem hiding this comment.
Pull Request Overview
The PR renames the core module and files from flash_dma to flash_dmattn, reorganizes CUDA source paths into an instantiations folder, and improves the kernel generator CLI by updating its description and adding a default output directory.
- Rename all occurrences of
flash_dmatoflash_dmattn, including package name, env vars, imports, and extension names - Move CUDA kernel source files into
csrc/src/instantiations/for better organization - Update
generate_kernels.pyto default--output_dirtoinstantiationsand adjust its description
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| setup.py | Updated package/env var names and extension identifiers |
| csrc/src/generate_kernels.py | Changed CLI description and added default output directory |
| benchmarks/benchmark_forward_equivalence.py | Renamed imports and function references for CUDA/Triton |
Comments suppressed due to low confidence (2)
benchmarks/benchmark_forward_equivalence.py:36
- In the Triton import block,
flash_dmattn_funcis only defined in the except clause; on successful import you should alias or assign the imported function/module toflash_dmattn_funcso it's always defined before it's used.
print(f"❌ Failed to import flash_dmattn_triton: {e}")
csrc/src/generate_kernels.py:108
- The help text says it will default to the current directory, but the actual default is now
instantiations; update this message to reflect the new default folder.
help="Where to generate the kernels "
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.
Update module references, import statements, and function names to use the more descriptive
flash_dmattnnaming convention. Enhance the kernel generator by changing the description and adding a default output directory for better usability. Reorganize source file paths for improved code organization. Maintain consistency across CUDA and Triton implementations.