[pytorch][triton] Enabling TMA for flex-attention for supported device types #153662
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
…e types Summary: Currently flex-attention defaults to `USE_TMA=False`. We can enable TMA on devices which support it based on `has_triton_tma_device`. Test Plan: # Tritonbench results ``` buck2 run mode/opt //pytorch/tritonbench:run -- --op flex_attention --use-tma --mod-type all . . . (B, Hq, M, Hkv, N, D) | Mask Type compiled-latency compiled-tflops - USE_TMA = False (8, 16, 128, 16, 128, 128) | noop 0.027936 (±5.61%) 38.5859 (8, 16, 128, 16, 128, 128) | causal 0.017760 (±3.42%) 60.6946 (8, 16, 128, 16, 128, 128) | rel 0.028384 (±5.07%) 37.9769 (8, 16, 128, 16, 128, 128) | head_bias 0.027712 (±4.27%) 38.8978 (8, 16, 128, 16, 128, 128) | alibi 0.017920 (±3.21%) 60.1527 - USE_TMA = True (8, 16, 128, 16, 128, 128) | noop 0.025632 (±5.74%) 42.0543 (8, 16, 128, 16, 128, 128) | causal 0.015328 (±3.97%) 70.3246 (8, 16, 128, 16, 128, 128) | rel 0.025824 (±4.96%) 41.7416 (8, 16, 128, 16, 128, 128) | head_bias 0.025472 (±4.90%) 42.3185 (8, 16, 128, 16, 128, 128) | alibi 0.015392 (±3.74%) 70.0322 ``` Differential Revision: D74841543
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/153662
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 83d37fd with merge base ea17cd0 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
This pull request was exported from Phabricator. Differential Revision: D74841543 |
|
Can you also do a perf bench w/ larger sequen lengths, I am curious Copying over comments: |
Skylion007
approved these changes
May 16, 2025
|
https://www.internalfb.com/intern/paste/P1813676625/ Some failing tests |
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.
Summary:
Currently flex-attention defaults to
USE_TMA=False.We can enable TMA on devices which support it based on
has_triton_tma_device.Test Plan:
Tritonbench results
Differential Revision: D74841543
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov