[ET-VK] Statically quantized convolutions #14647
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
## Changes This diff adds implementations for quantized convolution under the following quantization conditions: * activations statically quantized to 8-bit with per tensor scale and zero point * weights quantized to 8-bit with per channel scales * outputs statically quantized to 8-bit with per tensor scale and zero point 3 different implementations are added, which are selected between based on the input conditions. The first is an direct convolution shader which uses the quantized int8 input directly. The second is an im2col variant, which computes the convolution via a gemm like algorithm by first applying an im2col tranformation on the input tensor. Finally, a specialized implementation is added for depthwise convolutions. Differential Revision: [D83437827](https://our.internmc.facebook.com/intern/diff/D83437827/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14647
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 2 Cancelled JobsAs of commit e24e6ee with merge base 049c9fc ( 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
This was referenced Sep 28, 2025
This PR needs a
|
## Changes This diff adds implementations for quantized convolution under the following quantization conditions: * activations statically quantized to 8-bit with per tensor scale and zero point * weights quantized to 8-bit with per channel scales * outputs statically quantized to 8-bit with per tensor scale and zero point 3 different implementations are added, which are selected between based on the input conditions. The first is an direct convolution shader which uses the quantized int8 input directly. The second is an im2col variant, which computes the convolution via a gemm like algorithm by first applying an im2col tranformation on the input tensor. Finally, a specialized implementation is added for depthwise convolutions. Differential Revision: [D83437827](https://our.internmc.facebook.com/intern/diff/D83437827/) [ghstack-poisoned]
andreanicastro
had a problem deploying
to
cherry-pick-bot
GitHub Actions
Failure
andreanicastro
approved these changes
Sep 29, 2025
## Changes This diff adds implementations for quantized convolution under the following quantization conditions: * activations statically quantized to 8-bit with per tensor scale and zero point * weights quantized to 8-bit with per channel scales * outputs statically quantized to 8-bit with per tensor scale and zero point 3 different implementations are added, which are selected between based on the input conditions. The first is an direct convolution shader which uses the quantized int8 input directly. The second is an im2col variant, which computes the convolution via a gemm like algorithm by first applying an im2col tranformation on the input tensor. Finally, a specialized implementation is added for depthwise convolutions. Differential Revision: [D83437827](https://our.internmc.facebook.com/intern/diff/D83437827/) [ghstack-poisoned]
manuelcandales
approved these changes
Sep 29, 2025
facebook-github-bot
merged commit 32b94aa
into
gh/SS-JIA/332/base
125 of 132 checks passed
facebook-github-bot
temporarily deployed
to
cherry-pick-bot
GitHub Actions
Inactive
SS-JIA
pushed a commit
that referenced
this pull request
Sep 29, 2025
This PR was created by the merge bot to help merge the original PR into the main branch. ghstack PR number: #14647 by @SS-JIA ^ 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/SS-JIA/332/base ghstack PR head: https://github.com/pytorch/executorch/tree/gh/SS-JIA/332/head Merge bot PR base: https://github.com/pytorch/executorch/tree/main Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/SS-JIA/332/orig Differential Revision: [D83437827](https://our.internmc.facebook.com/intern/diff/D83437827/) @diff-train-skip-merge Co-authored-by: ssjia <ssjia@devvm26340.ftw0.facebook.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):
Changes
This diff adds implementations for quantized convolution under the following quantization conditions:
3 different implementations are added, which are selected between based on the input conditions.
The first is an direct convolution shader which uses the quantized int8 input directly.
The second is an im2col variant, which computes the convolution via a gemm like algorithm by first applying an im2col tranformation on the input tensor.
Finally, a specialized implementation is added for depthwise convolutions.
Differential Revision: D83437827