Using buffer for weight tensors for quantized mat mul op. #15990
+43
−31
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/15990
Note: Links to docs will display an error until the docs builds have been completed. ❗ 2 Active SEVsThere are 2 currently active SEVs. If your PR is affected, please view them below:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
|
@trivedivivek has exported this pull request. If you are a Meta employee, you can view the originating Diff in D87911255. |
trivedivivek
added
the
release notes: vulkan
yipjustin
approved these changes
Nov 26, 2025
trivedivivek
added a commit
to trivedivivek/executorch
that referenced
this pull request
Nov 26, 2025
) Summary: This change affects the performance and memory usage of the quantized matrix multiplication operation in the Executorch Vulkan backend. By using a buffer for weight tensors, the operation may become more efficient and use less memory, especially for large matrices. Reviewed By: yipjustin Differential Revision: D87911255
trivedivivek
force-pushed
the
export-D87911255
branch
2 times, most recently
from
fe35988 to
88fe2f2
Compare
trivedivivek
added a commit
to trivedivivek/executorch
that referenced
this pull request
Nov 26, 2025
) Summary: This change affects the performance and memory usage of the quantized matrix multiplication operation in the Executorch Vulkan backend. By using a buffer for weight tensors, the operation may become more efficient and use less memory, especially for large matrices. Reviewed By: yipjustin Differential Revision: D87911255
…r. (pytorch#15988) Summary: This diff optimizes the performance of the quantized linear matrix multiplication operation by using a 4x3 tiled shader, which performs slightly better than the previous implementation. Reviewed By: yipjustin Differential Revision: D87902847
) Summary: This change affects the performance and memory usage of the quantized matrix multiplication operation in the Executorch Vulkan backend. By using a buffer for weight tensors, the operation may become more efficient and use less memory, especially for large matrices. Reviewed By: yipjustin Differential Revision: D87911255
trivedivivek
force-pushed
the
export-D87911255
branch
from
88fe2f2 to
fd99f4e
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Summary: This change affects the performance and memory usage of the quantized matrix multiplication operation in the Executorch Vulkan backend. By using a buffer for weight tensors, the operation may become more efficient and use less memory, especially for large matrices.
Differential Revision: D87911255