Remove unused mask operand from Cadence quantized softmax - #21606
Remove unused mask operand from Cadence quantized softmax#21606DrJessop wants to merge 2 commits into
mask operand from Cadence quantized softmax#21606Conversation
Summary: The kernel supports masking, need to support it in the reference as well. Reviewed By: aliafzal Differential Revision: D114671838
Summary: Remove the unused `mask` operand from the quantized softmax schemas, fake and reference implementations, generic and DLA kernel APIs, and tests. Stop materializing a placeholder packed mask in the softmax quantizer pattern. Preserve position-based causal masking through `mask_type` and `pos`, and remove the dead DLA mask buffer. Differential Revision: D114967226
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21606
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 2 New Failures, 3 Cancelled JobsAs of commit d841c77 with merge base bb6b99a ( NEW FAILURES - The following jobs have failed:
CANCELLED JOBS - The following jobs were cancelled. Please retry:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
|
@DrJessop has exported this pull request. If you are a Meta employee, you can view the originating Diff in D114967226. |
This PR needs a
|
1 participant
Summary: Remove the unused
maskoperand from the quantized softmax schemas, fake and reference implementations, generic and DLA kernel APIs, and tests. Stop materializing a placeholder packed mask in the softmax quantizer pattern. Preserve position-based causal masking throughmask_typeandpos, and remove the dead DLA mask buffer.Differential Revision: D114967226