[ROCm] Prevent accidental enablement of efficient attention. #133331
Conversation
…d_tensors_cuda ROCm's Efficient Attention (GPU kernel shared with FA) is more tolerancing about the inputs.
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/133331
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (2 Unrelated Failures)As of commit e094699 with merge base 89795da ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
BROKEN TRUNK - The following job failed but was present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@xinyazhang is this read for review? If so, can you please remove draft status |
Yes this is ready. I'll implement your suggestion and move it out of draft status. |
|
@pytorchbot merge |
Merge failedReason: This PR needs a If not, please add the To add a label, you can comment to pytorchbot, for example For more information, see Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot label "topic: not user facing" |
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
|
The merge job was canceled or timed out. This most often happen if two merge requests were issued for the same PR, or if merge job was waiting for more than 6 hours for tests to finish. In later case, please do not hesitate to reissue the merge command |
|
@pytorchbot merge -f "Unrelated CI failures. Critical fix needed for 2.4.1" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
|
@pytorchbot cherry-pick --onto release/2.4 -c critical |
Currently Efficient attention and Flash attention share the same set of GPU kernels on ROCM and have common limitations on head sizes. Fixes #132004 Pull Request resolved: #133331 Approved by: https://github.com/malfet, https://github.com/jithunnair-amd (cherry picked from commit 46ecc67)
Cherry picking #133331The cherry pick PR is at #134531 and it is recommended to link a critical cherry pick PR with an issue. The following tracker issues are updated: Details for Dev Infra teamRaised by workflow job |
[ROCm] Prevent accidental enablement of efficient attention. (#133331) Currently Efficient attention and Flash attention share the same set of GPU kernels on ROCM and have common limitations on head sizes. Fixes #132004 Pull Request resolved: #133331 Approved by: https://github.com/malfet, https://github.com/jithunnair-amd (cherry picked from commit 46ecc67) Co-authored-by: Xinya Zhang <Xinya.Zhang@amd.com>
…#134531) [ROCm] Prevent accidental enablement of efficient attention. (pytorch#133331) Currently Efficient attention and Flash attention share the same set of GPU kernels on ROCM and have common limitations on head sizes. Fixes pytorch#132004 Pull Request resolved: pytorch#133331 Approved by: https://github.com/malfet, https://github.com/jithunnair-amd (cherry picked from commit 46ecc67) Co-authored-by: Xinya Zhang <Xinya.Zhang@amd.com>
…#134531) (#1565) [ROCm] Prevent accidental enablement of efficient attention. (pytorch#133331) Currently Efficient attention and Flash attention share the same set of GPU kernels on ROCM and have common limitations on head sizes. Pull Request resolved: pytorch#133331 Approved by: https://github.com/malfet, https://github.com/jithunnair-amd (cherry picked from commit 46ecc67) Fixes pytorch#132004 Co-authored-by: pytorchbot <soumith+bot@pytorch.org>
…#133331) Currently Efficient attention and Flash attention share the same set of GPU kernels on ROCM and have common limitations on head sizes. Fixes pytorch#132004 Pull Request resolved: pytorch#133331 Approved by: https://github.com/malfet, https://github.com/jithunnair-amd
Currently Efficient attention and Flash attention share the same set of GPU
kernels on ROCM and have common limitations on head sizes.
Fixes #132004
cc @jeffdaily @sunway513 @jithunnair-amd @pruthvistony @ROCmSupport @dllehr-amd @jataylo @hongxiayang