[ONNX] Support fixed-capacity GroupQueryAttention cache#4534
Draft
[ONNX] Support fixed-capacity GroupQueryAttention cache#4534
Conversation
IanWood1
force-pushed
the
fix-constant-seq-length-onnx-gqa
branch
2 times, most recently
from
b3e6c84 to
1c69b44
Compare
IanWood1
force-pushed
the
fix-constant-seq-length-onnx-gqa
branch
2 times, most recently
from
cf09e66 to
00ecc03
Compare
IanWood1
changed the title
Signed-off-by: Ian Wood <ianwood@u.northwestern.edu>
IanWood1
force-pushed
the
fix-constant-seq-length-onnx-gqa
branch
from
00ecc03 to
18d6386
Compare
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 updates the GroupQueryAttention lowering to support the past_present_share_buffer style runtime behavior, where present_key/present_value have the same fixed-capacity cache type as past_key/past_value. This intentionally drops support for the non-past_present_share_buffer behavior in this lowering.
Supporting both (choosing at runtime) requires comparing the input cache size to the output cache size to determine which mode. For dynamic seqlen, this is not possible because there is no way to get the output KV cache size.
Also, this adds explicit support for rank 2 seqlens_k which is off spec but emitted by onnx exporters.