Reapply #11294 and #11295 (improve GLU test and implement using internal views to avoid copying) #11509
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
…nal views to avoid copying) These were reverted due to internal test failures. Sending this as an exported internal diff so that we can make sure we get internal signal. Original summary for #11294 (to make the GLU test input asymmetric): This way it will produce different results along each tested dim. Original summaryfor #11295: GLU requires slicing the input Tensor into two halves. Currently, we accomplish this by copying; ExecuTorch does not support views in general because it requires Tensors to be contiguous. However, nothing stops us from implementing [the ATen that uses views](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/GatedLinearUnit.cpp#L35) entirely internally to the op. To support this, I added `support_noncontiguous_tensors` as an optional template argument to BroadcastIndexesRange and plumbed it through to the elementwise_util functions as an optional SupportNonContiguousTensors parameter. Differential Revision: [D76311585](https://our.internmc.facebook.com/intern/diff/D76311585/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11509
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit 3d39332 with merge base ef1d2ff ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
swolchok
added a commit
that referenced
this pull request
Jun 10, 2025
…nal views to avoid copying) These were reverted due to internal test failures. Sending this as an exported internal diff so that we can make sure we get internal signal. Original summary for #11294 (to make the GLU test input asymmetric): This way it will produce different results along each tested dim. Original summaryfor #11295: GLU requires slicing the input Tensor into two halves. Currently, we accomplish this by copying; ExecuTorch does not support views in general because it requires Tensors to be contiguous. However, nothing stops us from implementing [the ATen that uses views](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/GatedLinearUnit.cpp#L35) entirely internally to the op. To support this, I added `support_noncontiguous_tensors` as an optional template argument to BroadcastIndexesRange and plumbed it through to the elementwise_util functions as an optional SupportNonContiguousTensors parameter. Differential Revision: [D76311585](https://our.internmc.facebook.com/intern/diff/D76311585/) ghstack-source-id: 289287609 Pull Request resolved: #11509
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D76311585 |
This PR needs a
|
lucylq
approved these changes
Jun 10, 2025
…d implement using internal views to avoid copying)" These were reverted due to internal test failures. Sending this as an exported internal diff so that we can make sure we get internal signal. Original summary for #11294 (to make the GLU test input asymmetric): This way it will produce different results along each tested dim. Original summaryfor #11295: GLU requires slicing the input Tensor into two halves. Currently, we accomplish this by copying; ExecuTorch does not support views in general because it requires Tensors to be contiguous. However, nothing stops us from implementing [the ATen that uses views](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/GatedLinearUnit.cpp#L35) entirely internally to the op. To support this, I added `support_noncontiguous_tensors` as an optional template argument to BroadcastIndexesRange and plumbed it through to the elementwise_util functions as an optional SupportNonContiguousTensors parameter. Differential Revision: [D76311585](https://our.internmc.facebook.com/intern/diff/D76311585/) [ghstack-poisoned]
swolchok
added a commit
that referenced
this pull request
Jun 10, 2025
…nal views to avoid copying) Pull Request resolved: #11509 These were reverted due to internal test failures. Sending this as an exported internal diff so that we can make sure we get internal signal. Original summary for #11294 (to make the GLU test input asymmetric): This way it will produce different results along each tested dim. Original summaryfor #11295: GLU requires slicing the input Tensor into two halves. Currently, we accomplish this by copying; ExecuTorch does not support views in general because it requires Tensors to be contiguous. However, nothing stops us from implementing [the ATen that uses views](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/GatedLinearUnit.cpp#L35) entirely internally to the op. To support this, I added `support_noncontiguous_tensors` as an optional template argument to BroadcastIndexesRange and plumbed it through to the elementwise_util functions as an optional SupportNonContiguousTensors parameter. Differential Revision: [D76311585](https://our.internmc.facebook.com/intern/diff/D76311585/) ghstack-source-id: 289429482
|
This pull request was exported from Phabricator. Differential Revision: D76311585 |
…using internal views to avoid copying)" These were reverted due to internal test failures. Sending this as an exported internal diff so that we can make sure we get internal signal. Original summary for #11294 (to make the GLU test input asymmetric): This way it will produce different results along each tested dim. Original summaryfor #11295: GLU requires slicing the input Tensor into two halves. Currently, we accomplish this by copying; ExecuTorch does not support views in general because it requires Tensors to be contiguous. However, nothing stops us from implementing [the ATen that uses views](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/GatedLinearUnit.cpp#L35) entirely internally to the op. To support this, I added `support_noncontiguous_tensors` as an optional template argument to BroadcastIndexesRange and plumbed it through to the elementwise_util functions as an optional SupportNonContiguousTensors parameter. Differential Revision: [D76311585](https://our.internmc.facebook.com/intern/diff/D76311585/) [ghstack-poisoned]
swolchok
added a commit
that referenced
this pull request
Jun 10, 2025
…nal views to avoid copying) Pull Request resolved: #11509 These were reverted due to internal test failures. Sending this as an exported internal diff so that we can make sure we get internal signal. Original summary for #11294 (to make the GLU test input asymmetric): This way it will produce different results along each tested dim. Original summaryfor #11295: GLU requires slicing the input Tensor into two halves. Currently, we accomplish this by copying; ExecuTorch does not support views in general because it requires Tensors to be contiguous. However, nothing stops us from implementing [the ATen that uses views](https://github.com/pytorch/pytorch/blob/main/aten/src/ATen/native/GatedLinearUnit.cpp#L35) entirely internally to the op. To support this, I added `support_noncontiguous_tensors` as an optional template argument to BroadcastIndexesRange and plumbed it through to the elementwise_util functions as an optional SupportNonContiguousTensors parameter. ghstack-source-id: 289432903 Differential Revision: [D76311585](https://our.internmc.facebook.com/intern/diff/D76311585/)
|
This pull request was exported from Phabricator. Differential Revision: D76311585 |
facebook-github-bot
merged commit 362929b
into
gh/swolchok/451/base
95 of 98 checks passed
facebook-github-bot
temporarily deployed
to
cherry-pick-bot
GitHub Actions
Inactive
swolchok
added a commit
that referenced
this pull request
Jun 11, 2025
…nal views to avoid copying) (#11539) This PR was created by the merge bot to help merge the original PR into the main branch. ghstack PR number: #11509 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/451/base ghstack PR head: https://github.com/pytorch/executorch/tree/gh/swolchok/451/head Merge bot PR base: https://github.com/pytorch/executorch/tree/main Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/swolchok/451/orig @diff-train-skip-merge Co-authored-by: Scott Wolchok <[email protected]>
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.
Stack from ghstack (oldest at bottom):
These were reverted due to internal test failures. Sending this as an exported internal diff so that we can make sure we get internal signal.
Original summary for #11294 (to make the GLU test input asymmetric):
This way it will produce different results along each tested dim.
Original summaryfor #11295:
GLU requires slicing the input Tensor into two halves. Currently, we
accomplish this by copying; ExecuTorch does not support views in general
because it requires Tensors to be contiguous. However, nothing stops us
from implementing the ATen that uses
views
entirely internally to the op.
To support this, I added
support_noncontiguous_tensorsas an optionaltemplate argument to BroadcastIndexesRange and plumbed it through to the
elementwise_util functions as an optional SupportNonContiguousTensors
parameter.
Differential Revision: D76311585