Repository navigation
Conversation
Configure native SDPA through quantizer annotations and decomposition provenance instead of requiring model-specific attention arithmetic. Support A16 attention and experimental A8 QK/AV, deriving INT32 QK scales from symmetric operands and retaining A16 masked softmax. Keep calibrated integer cache storage separate from attention policy. Retain the experimental Vela-specific INT16 variable path without a model-specific SDPA source transform. Cover quantizer selection, masks, GQA, accumulator bounds, and delegated mutable-buffer ownership. AI-assisted-by: Codex Change-Id: Ia0d7d8113e3d0f6aa6b625a121329c86d7b6c539 Signed-off-by: Xingguo Li <xingguo.li@arm.com>
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/25033
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 217 PendingAs of commit ab23fc2 with merge base 765e973 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
xingguo01
added
release notes: arm
This branch has not been deployed
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Configure native SDPA through quantizer annotations and decomposition provenance instead of requiring model-specific attention arithmetic. Support A16 attention and experimental A8 QK/AV, deriving INT32 QK scales from symmetric operands and retaining A16 masked softmax.
Keep calibrated integer cache storage separate from attention policy. Retain the experimental Vela-specific INT16 variable path without a model-specific SDPA source transform. Cover quantizer selection, masks, GQA, accumulator bounds, and delegated mutable-buffer ownership.
AI-assisted-by: Codex
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani