Improve Shape Inference for GQA #24143
Merged
+23
−0
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
tianleiwu
reviewed
Mar 24, 2025
tianleiwu
reviewed
Mar 25, 2025
tianleiwu
approved these changes
Mar 25, 2025
|
/azp run Windows ARM64 QNN CI Pipeline,Windows x64 QNN CI Pipeline,Windows GPU Doc Gen CI Pipeline,ONNX Runtime Web CI Pipeline,Win_TRT_Minimal_CUDA_Test_CI,Linux CPU CI Pipeline,Linux CPU Minimal Build E2E CI Pipeline,Linux GPU CI Pipeline,Linux GPU TensorRT CI Pipeline,Linux OpenVINO CI Pipeline |
|
/azp run Linux QNN CI Pipeline,onnxruntime-binary-size-checks-ci-pipeline,Big Models,Linux Android Emulator QNN CI Pipeline,Android CI Pipeline,iOS CI Pipeline,ONNX Runtime React Native CI Pipeline,Linux DNNL CI Pipeline,Linux MIGraphX CI Pipeline,Linux ROCm CI Pipeline |
|
Azure Pipelines successfully started running 8 pipeline(s). |
|
Azure Pipelines successfully started running 10 pipeline(s). |
aciddelgado
reviewed
Mar 25, 2025
quic-zhaoxul
pushed a commit
to CodeLinaro/onnxruntime
that referenced
this pull request
Apr 17, 2025
### Description <!-- Describe your changes. --> For GroupQueryAttention op, if the input total_sequence_length is a constant, we can infer the shape of output present_key/present_value `(batch_size, kv_num_heads, present_sequence_length, head_size)`. https://github.com/microsoft/onnxruntime/blob/5ed900e9712ce2f02e40c15b945d18453d1960d8/onnxruntime/contrib_ops/cpu/bert/group_query_attention_helper.h#L185 We know that from CPU EP, `present_sequence_length = max(past_sequence_length, total_sequence_length)`, and `batch_size, kv_num_heads, head_size` are the same as past_key/past_value. This inference is very important for WebNN EP, because WebNN only supports GQA for `present_sequence_length == past_sequence_length` and requires static shape for graph compilation. ### Motivation and Context <!-- - Why is this change required? What problem does it solve? - If it fixes an open issue, please link to the issue here. -->
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.
Description
For GroupQueryAttention op, if the input total_sequence_length is a constant, we can infer the shape of output present_key/present_value
(batch_size, kv_num_heads, present_sequence_length, head_size).onnxruntime/onnxruntime/contrib_ops/cpu/bert/group_query_attention_helper.h
Line 185 in 5ed900e
We know that from CPU EP,
present_sequence_length = max(past_sequence_length, total_sequence_length), andbatch_size, kv_num_heads, head_sizeare the same as past_key/past_value.This inference is very important for WebNN EP, because WebNN only supports GQA for
present_sequence_length == past_sequence_lengthand requires static shape for graph compilation.Motivation and Context