Removes no-topk CUDA implementation from benchmarks #65
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
Eliminates the dynamic mask attention CUDA implementation without topk computation to simplify benchmark comparisons and reduce code complexity. Updates test configurations to use head dimension of 128 instead of 32 for more realistic performance testing scenarios. Adjusts benchmark output formatting to accommodate the reduced number of implementations being compared.
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
This PR removes the no-topk CUDA implementation from the benchmarks to simplify code and comparisons. The changes update test configurations to use a more realistic head dimension of 128 instead of 32, and adjust the benchmark output formatting to accommodate fewer implementations.
- Removes the
dynamic_mask_attention_cuda_no_topkfunction and all related benchmarking code - Updates test configurations to use head dimension 128 instead of 32 for more realistic scenarios
- Adjusts benchmark output formatting by reducing table width and removing no-topk columns
| flash_avg = time_avgs.get('flash', float('inf')) | ||
|
|
||
| for impl_key in ['cuda', 'no_topk', 'triton', 'flex']: | ||
| for impl_key in ['cuda', 'triton', 'flex']: |
There was a problem hiding this comment.
The hardcoded list of implementation keys should be extracted to a constant or derived from the implementations dictionary to avoid maintenance issues when adding or removing implementations.
Suggested change
| for impl_key in ['cuda', 'triton', 'flex']: | |
| for impl_key in [key for key in implementations.keys() if key != 'flash']: |
Removes flash attention backend specification to use default SDPA behavior and enables attention mask usage. Comments out most benchmark configurations to focus testing on window size variations, reducing benchmark execution time while maintaining core functionality testing.
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.
Eliminates the dynamic mask attention CUDA implementation without topk computation to simplify benchmark comparisons and reduce code complexity.
Updates test configurations to use head dimension of 128 instead of 32 for more realistic performance testing scenarios.
Adjusts benchmark output formatting to accommodate the reduced number of implementations being compared.