Arm backend: Add INT16 support to rescale operation #13802
Merged
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
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) [ghstack-poisoned]
This was referenced Aug 29, 2025
This was referenced Aug 29, 2025
Ninja91
added a commit
that referenced
this pull request
Aug 29, 2025
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) ghstack-source-id: 304555411 Pull Request resolved: #13802
meta-cla
bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
zingo
added
ciflow/trunk
module: arm
zingo
changed the title
digantdesai
reviewed
Aug 29, 2025
digantdesai
requested changes
Aug 29, 2025
There was a problem hiding this comment.
I didn't see torch ops like tests for TOSA dialect ops, perhaps we should add it here. @per what do you think?
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) cc digantdesai freddan80 per zingo oscarandersson8218 [ghstack-poisoned]
Ninja91
added a commit
that referenced
this pull request
Sep 4, 2025
Pull Request resolved: #13802 Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. ghstack-source-id: 304555411 @exported-using-ghexport Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/)
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
digantdesai
approved these changes
Sep 5, 2025
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) cc digantdesai freddan80 per zingo oscarandersson8218 [ghstack-poisoned]
Ninja91
added a commit
that referenced
this pull request
Sep 6, 2025
Pull Request resolved: #13802 Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. ghstack-source-id: 308021436 @exported-using-ghexport Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/)
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) cc digantdesai freddan80 per zingo oscarandersson8218 [ghstack-poisoned]
Ninja91
added a commit
that referenced
this pull request
Sep 6, 2025
Pull Request resolved: #13802 Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. ghstack-source-id: 308024305 @exported-using-ghexport Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/)
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) cc digantdesai freddan80 per zingo oscarandersson8218 [ghstack-poisoned]
Ninja91
added a commit
that referenced
this pull request
Sep 10, 2025
Pull Request resolved: #13802 Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. ghstack-source-id: 308024305 @exported-using-ghexport Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/)
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) cc digantdesai freddan80 per zingo oscarandersson8218 [ghstack-poisoned]
Ninja91
added a commit
that referenced
this pull request
Sep 10, 2025
Pull Request resolved: #13802 Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. ghstack-source-id: 308860606 @exported-using-ghexport Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/)
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) cc digantdesai freddan80 per zingo oscarandersson8218 [ghstack-poisoned]
Ninja91
added a commit
that referenced
this pull request
Sep 10, 2025
Pull Request resolved: #13802 Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. ghstack-source-id: 308928949 @exported-using-ghexport Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/)
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/) cc digantdesai freddan80 per zingo oscarandersson8218 [ghstack-poisoned]
Ninja91
added a commit
that referenced
this pull request
Sep 11, 2025
Pull Request resolved: #13802 Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend. This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations. Changes: - Add INT16 dtype validation support in op_rescale.py - Enable rescale operations for 16A8W quantization configuration The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline. ghstack-source-id: 308986675 @exported-using-ghexport Differential Revision: [D80513725](https://our.internmc.facebook.com/intern/diff/D80513725/)
|
This pull request was exported from Phabricator. Differential Revision: D80513725 |
facebook-github-bot
merged commit 8c3c565
into
gh/Ninja91/15/base
291 of 303 checks passed
facebook-github-bot
had a problem deploying
to
cherry-pick-bot
GitHub Actions
Failure
facebook-github-bot
had a problem deploying
to
cherry-pick-bot
GitHub Actions
Failure
|
@Ninja91 You need to land this in main cc @digantdesai (reviewer) @lucylq (oncall) |
lucylq
pushed a commit
that referenced
this pull request
Sep 15, 2025
Differential Revision: D80513725 Pull Request resolved: #13802
lucylq
added a commit
that referenced
this pull request
Sep 15, 2025
Differential Revision: D80513725 Pull Request resolved: #13802 #13802 (comment) failed to cp to main Co-authored-by: Nitin Jain <jainnitin@meta.com>
StrycekSimon
pushed a commit
to nxp-upstream/executorch
that referenced
this pull request
Sep 23, 2025
Differential Revision: D80513725 Pull Request resolved: pytorch#13802 pytorch#13802 (comment) failed to cp to main Co-authored-by: Nitin Jain <jainnitin@meta.com>
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):
Add INT16 support for RequantizeNode rescale operations in ExecutorTorch ARM backend.
This follows the pattern established for linear, mul, sigmoid, tanh, slice, view/transpose, cat, and FCNode operations, extending int16 support to RequantizeNode rescale operations.
Changes:
The 16A8W configuration uses 16-bit activations with 8-bit weights, enabling higher precision for activations while maintaining weight efficiency. RequantizeNode rescale operations are essential for proper quantization scaling in the 16A8W pipeline.
Differential Revision: D80513725
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218