[DRAFT] Use mask instead of cond for attention conditional logic #7536
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/7536
Note: Links to docs will display an error until the docs builds have been completed. ❌ 4 New FailuresAs of commit 220aec8 with merge base ca32105 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
This PR needs a
|
jackzhxng
commented
Jan 7, 2025
| k, v = self.kv_cache.update(k, v) | ||
|
|
||
| output = self._sdpa(q, k, v, b, s_x, mask=mask) | ||
| output = self._sdpa(q, k, v, b, s_x) |
Contributor
Author
There was a problem hiding this comment.
Note to self: the refactor removed the mask arg, make sure to add it back
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Outdated suggestions cannot be applied.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
Use masking to avoid resorting to
torch.cond, which prevents us from mutating inside branches thereby forcing us to clone the kv cache and create lots of unnecessary copies.Also gets past the current limitation that the partitioners don't automatically recursively partition conditional subgraphs, allowing us to directly partition Llama 3.2 MM with XNNPack.
Llama 3.2 MM comparison against XNNPack + KV cache + custom SDPA
Test plan
Rely on existing regression tests (test_attention and test_kv_cache - about to be merged), which have adequate coverage over kv cache and multi-head attention edge cases.