-
Notifications
You must be signed in to change notification settings - Fork 650
[xnn update prep] deprecate sdpa #11506
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
Stack from ghstack (oldest at bottom): |
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11506
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: ❌ 4 New Failures, 7 Pending, 1 Unrelated FailureAs of commit 6e7028e with merge base d4cc258 ( NEW FAILURES - The following jobs have failed:
FLAKY - The following job failed but was likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
@@ -2097,7 +2061,6 @@ DefineNodeFunc getDefineNodeFunc(fb_xnnpack::XNodeUnion nodeType) { | |||
_DEFINE(Concatenate4) | |||
_DEFINE(Concatenate5) | |||
_DEFINE(StaticSlice) | |||
_DEFINE(ScaledDotProductAttention) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Not updating schema for marking deprecated? XNNScaledDotProductAttention
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
so XNNPACK removed the operator from their codebase, so for next update we need to delete. I can mark the operator in the schema as deprecated though.
@@ -1,111 +0,0 @@ | |||
# Copyright (c) Meta Platforms, Inc. and affiliates. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should we continue partitioning this and lower it as decomposed?
Else this will be a BC breaking change.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
So ever since my new partitioner, SDPA has not been delegated. I assumed that no models in production or since h ave really used any sdpa implementation (also because i heard our sdpa is slow). so I believe this is safe to remove. I can import it internally to make sure
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
What I was thinking was we should handle decomp of SDPA inside XNNPACK AoT so that we don't regress for current perf. Lowering pieces is OK but can cause unexpected perf drops with slight changes.
And I guess when the constraint fails it will get decomp and we will lower it in pieces anyway, right?
@mcr229 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
@mcr229 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
@mcr229 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
@mcr229 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
ghstack-source-id: 479c94a ghstack-comment-id: 2957149191 Pull Request resolved: pytorch/executorch#11506
ghstack-source-id: 479c94a ghstack-comment-id: 2957149191 Pull Request resolved: pytorch/executorch#11506
ghstack-source-id: 929a8bc ghstack-comment-id: 2957149191 Pull Request resolved: pytorch#11506
@mcr229 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Stamping to unblock rebase, but we should revisit this under the larger task of replacing custom SDPA op with XNNPACK's in the future.
@@ -27,8 +27,6 @@ bool check_tensor_dtype( | |||
return executorch::runtime::tensor_is_floating_type(t); | |||
case SupportedTensorDtypes::INTB: | |||
return executorch::runtime::tensor_is_integral_type(t, true); | |||
case SupportedTensorDtypes::BOOL: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why is this change here? bad rebase?
This change seems to have accidentally reverted several of my changes due to (presumably) a bad rebase. Differential Revision: [D77559984](https://our.internmc.facebook.com/intern/diff/D77559984/) [ghstack-poisoned]
This change seems to have accidentally reverted several of my changes due to (presumably) a bad rebase. Differential Revision: [D77559984](https://our.internmc.facebook.com/intern/diff/D77559984/) [ghstack-poisoned]
This change seems to have accidentally reverted several of my changes due to (presumably) a bad rebase. Differential Revision: [D77559984](https://our.internmc.facebook.com/intern/diff/D77559984/) [ghstack-poisoned]
Pull Request resolved: #12121 This change seems to have accidentally reverted several of my changes due to (presumably) a bad rebase. ghstack-source-id: 293898720 @exported-using-ghexport Differential Revision: [D77559984](https://our.internmc.facebook.com/intern/diff/D77559984/)
Differential Revision: D77559984 Pull Request resolved: #12121
This PR was created by the merge bot to help merge the original PR into the main branch. ghstack PR number: #12121 by @swolchok ^ Please use this as the source of truth for the PR details, comments, and reviews ghstack PR base: https://github.com/pytorch/executorch/tree/gh/swolchok/487/base ghstack PR head: https://github.com/pytorch/executorch/tree/gh/swolchok/487/head Merge bot PR base: https://github.com/pytorch/executorch/tree/main Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/swolchok/487/orig @diff-train-skip-merge Co-authored-by: Scott Wolchok <[email protected]>
This PR was created by the merge bot to help merge the original PR into the main branch. ghstack PR number: pytorch#12121 by @swolchok ^ Please use this as the source of truth for the PR details, comments, and reviews ghstack PR base: https://github.com/pytorch/executorch/tree/gh/swolchok/487/base ghstack PR head: https://github.com/pytorch/executorch/tree/gh/swolchok/487/head Merge bot PR base: https://github.com/pytorch/executorch/tree/main Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/swolchok/487/orig @diff-train-skip-merge Co-authored-by: Scott Wolchok <[email protected]>
Differential Revision: D77265464