Add ONNX Runtime GQA-style SDPA benchmark#18716
Open
kimishpatel wants to merge 1 commit intogh/kimishpatel/236/basefrom
Open
Add ONNX Runtime GQA-style SDPA benchmark#18716kimishpatel wants to merge 1 commit intogh/kimishpatel/236/basefrom
kimishpatel wants to merge 1 commit intogh/kimishpatel/236/basefrom
Conversation
Ports the attention algorithm from onnxruntime's gqa_attention_base.h to enable direct performance comparison against ET's flash attention and the existing standard SDPA benchmark. Key differences from standard SDPA: scale baked into GEMM alpha (saves a scaling pass), scores buffer padded to max_seq_len columns (matching ONNX's present_buffer_sequence_length), narrow softmax over valid causal window only (zeros elsewhere, skips exp on masked positions), and output in [B,S,Hq,D] with stride Hq*D matching ONNX's interleaved output format. Validation tests confirm ONNX GQA matches ET custom_sdpa_out within float32 tolerance. Authored with Claude. Differential Revision: [D99677677](https://our.internmc.facebook.com/intern/diff/D99677677/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18716
Note: Links to docs will display an error until the docs builds have been completed. ❌ 127 New FailuresAs of commit 41e6c73 with merge base fb1618e ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Apr 6, 2026
Open
This PR needs a
|
meta-cla
bot
added
the
CLA Signed
Contributor
Author
|
submitted by accident, not meant to land immedidately |
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.
Stack from ghstack (oldest at bottom):
Ports the attention algorithm from onnxruntime's gqa_attention_base.h
to enable direct performance comparison against ET's flash attention
and the existing standard SDPA benchmark.
Key differences from standard SDPA: scale baked into GEMM alpha
(saves a scaling pass), scores buffer padded to max_seq_len columns
(matching ONNX's present_buffer_sequence_length), narrow softmax
over valid causal window only (zeros elsewhere, skips exp on masked
positions), and output in [B,S,Hq,D] with stride Hq*D matching
ONNX's interleaved output format.
Validation tests confirm ONNX GQA matches ET custom_sdpa_out within
float32 tolerance.
Authored with Claude.
Differential Revision: D99677677