Conversation
Pull Request resolved: #14130 ## Context As title; optimize the SDPA operator by introducing shaders to perform the operation in 3 steps: 1. Compute attention weights, multiplying QT x K_cache, and applying scale and mask 2. Compute softmax normalization of computed attention weights 3. Compute final output by multiplying attention weights with V cache This new implementation is much more efficient than the existing one, which performed slicing, repeat_interleave, and transposition of projected and cache tensors as separate steps. The fusion of scale and mask with the computation of attention weights also allows for the computation of elements within the mask region to be skipped. ## Impact Decode latency for LLMs is much improved. For llama 3.2 3B generating ~250 tokens, decode latency increases from ~15 tok/s to ~21.5 tok/s ghstack-source-id: 308660072 @exported-using-ghexport Differential Revision: [D82053493](https://our.internmc.facebook.com/intern/diff/D82053493/)
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14139
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 141 PendingAs of commit 5252297 with merge base 245630a ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
SS-JIA
approved these changes
Sep 10, 2025
This PR needs a
|
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR was created by the merge bot to help merge the original PR into the main branch.
ghstack PR number: #14130 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/324/base
ghstack PR head: https://github.com/pytorch/executorch/tree/gh/SS-JIA/324/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/324/orig
@diff-train-skip-merge