Reduce CPU Concat ops in Gemma4 audio encoder - #270
Merged
Conversation
Replace dynamic Shape+Concat shape construction with static value_ints in Reshape ops. ONNX Reshape treats 0 as 'copy from input dim', eliminating the need for runtime Shape extraction. Changes: - Q/K/V reshape [B, T, H*D] → [B, T, H, D]: use [0, 0, H, D] - Output reshape [B, T, H, D] → [B, T, H*D]: use [0, 0, -1] - Subsample reshape [B, T', F', C] → [B, T', F'*C]: use [0, 0, -1] Reduces audio encoder CPU Concat nodes from 49 to 24 (51% reduction). Remaining Concats are from causal window mask construction that genuinely requires dynamic shapes. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: Justin Chu <justinchu@microsoft.com>
37 tasks
Performance Comparison
|
Contributor
There was a problem hiding this comment.
Pull request overview
This PR reduces CPU-side Concat nodes in the Gemma4 audio encoder graph by replacing dynamic Shape+Concat-constructed reshape shapes with static Constant(value_ints=...) shapes that use ONNX Reshape’s 0 sentinel (“copy from input dim”). This aligns with the PR goal of reducing CPU ops on CUDA EP.
Changes:
- Replace dynamic
Shape+Concatshape construction in ConvSubsampling flattening withReshape(..., Constant([0, 0, -1])). - Replace dynamic
Shape+Concatshape construction for Q/K/V and output context reshapes withReshape(..., Constant([0, 0, num_heads, head_dim]))andConstant([0, 0, -1]).
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
🏗️ Architecture Diff
No architecture changes detected. ✅ Legend: ⚪ No change · 🔵 Minor (attrs/inits) · 🟡 Moderate (nodes added/removed) · 🔴 Major (interface changed) |
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.
Replace dynamic Shape+Concat with static value_ints in Reshape. Reduces CPU Concat from 49 to 24 (51%).