Support sliding-window layers on the smart_mask cache ring - #21929
Open
YIWENX14 wants to merge 2 commits into
Open
Support sliding-window layers on the smart_mask cache ring#21929YIWENX14 wants to merge 2 commits into
YIWENX14 wants to merge 2 commits into
Conversation
Summary: Chunked prefill of a sliding-window (local) attention layer let every query row see more keys than the window allows, so multi-chunk prefill did not match single-chunk prefill. This makes the cache region of the attention mask per-query-row in `StaticAttentionIOManager`. Reviewed By: billmguo Differential Revision: D116080453
Summary: Sliding-window (local) attention layers previously worked only with the `shift_pointer` KV cache layout. This makes them work with `smart_mask` as well, so chunked prefill matches single-chunk prefill under either layout. Differential Revision: D116524534
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21929
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 1 Unclassified FailureAs of commit b381b90 with merge base ed2d9ec ( NEW FAILURES - The following jobs have failed:
UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Contributor
|
@YIWENX14 has exported this pull request. If you are a Meta employee, you can view the originating Diff in D116524534. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary: Sliding-window (local) attention layers previously worked only with the
shift_pointerKV cache layout. This makes them work withsmart_maskas well, so chunked prefill matches single-chunk prefill under either layout.Differential Revision: D116524534