[SDPA] Add testing to ensure stride order exactly matches #152894
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/152894
Note: Links to docs will display an error until the docs builds have been completed. This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
Update mem eff striding |
drisspg
commented
May 6, 2025
test/test_transformers.py
Outdated
| @@ -2469,6 +2469,73 @@ def test_cudnn_attention_different_dk_dv(self, device): | |||
|
|
|||
| self.assertEqual(actual.contiguous(), math_ref.contiguous().to(dtype), atol=1e-3, rtol=1e-2) | |||
|
|
|||
| @unittest.skipIf(not PLATFORM_SUPPORTS_FUSED_ATTENTION, "Fused SDPA was not built for this system") | |||
| @parametrize("backend", PLATFORM_SPECIFIC_SDPA, name_fn=lambda x: x.name) | |||
| @parametrize("compile_mode", ["eager", "inductor"]) | |||
There was a problem hiding this comment.
So locally all test pass but that doesn't seem possible, since i didnt even update the meta for mem eff yet
i wonder if there are too many recompiles and then just falls back to eager..
eellison
reviewed
May 6, 2025
test/test_transformers.py
Outdated
| if compile_mode == "inductor": | ||
| run_sdpa = torch.compile(run_sdpa, backend="inductor", fullgraph=True) | ||
|
|
||
| with sdpa_kernel(backends=[backend]): |
There was a problem hiding this comment.
if compile_mode == "eager", can you enable CrossRefFakeMode ?
Skylion007
reviewed
May 7, 2025
Skylion007
reviewed
May 7, 2025
| grad_q = at::empty(query.sizes(), query.options()); | ||
| grad_k = at::empty(key.sizes(), key.options()); | ||
| grad_v = at::empty(value.sizes(), value.options()); | ||
| grad_q = at::empty_like(query); |
There was a problem hiding this comment.
Pretty sure flash attention used to have the same bug, I guess it was copied and pasted from here and never fixed here.
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 ca
304E
nnot 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.
Stack from ghstack (oldest at bottom):
Currently results
TODO update meta before land for mem eff
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov