Update QMoE emission for ORT raw-weight prepack support - #460
Open
justinchuby with Copilot wants to merge 2 commits into
Open
Update QMoE emission for ORT raw-weight prepack support#460justinchuby with Copilot wants to merge 2 commits into
justinchuby with Copilot wants to merge 2 commits into
Conversation
|
|
Emit the new weights_prepacked=0 QMoE attribute for raw int expert weights now that ORT can prepack them internally. Add focused assertions for both component-built and rewrite-fused QMoE nodes. Signed-off-by: GitHub <noreply@github.com> Co-authored-by: justinchuby <11205048+justinchuby@users.noreply.github.com>
Copilot
AI
changed the title
[WIP] Update MoE op after microsoft/onnxruntime#28749 lands
Update QMoE emission for ORT raw-weight prepack support
Aug 7, 2026
justinchuby
marked this pull request as ready for review
August 7, 2026 05:48
Performance Comparison
|
Contributor
There was a problem hiding this comment.
Pull request overview
Aligns Mobius’s com.microsoft::QMoE emission with the updated ORT QMoE ABI by explicitly opting into raw (non-prepacked) int4/int8 expert weights, preventing ORT CUDA from misinterpreting emitted raw tensors as prepacked.
Changes:
- Set
quant_type="int"andweights_prepacked=0when emitting fusedQMoEfromMoELayer. - Apply the same ABI attributes when rewriting dense-fallback MoE subgraphs into a single
QMoEnode. - Add regression assertions in both component and rewrite-rule tests to lock in the ABI attributes (Fixes #348).
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| src/mobius/rewrite_rules/_qmoe_fusion.py | Adds quant_type / weights_prepacked attributes to rewritten fused QMoE nodes. |
| src/mobius/rewrite_rules/_qmoe_fusion_test.py | Asserts the rewritten QMoE node includes the new ABI attributes. |
| src/mobius/components/_moe.py | Adds quant_type="int" and weights_prepacked=0 to MoELayer’s op.QMoE(...) emission. |
| src/mobius/components/_moe_test.py | Asserts MoELayer-emitted QMoE nodes include the new ABI attributes. |
🏗️ 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.
ORT QMoE now supports raw int4/int8 expert weights via
weights_prepacked=0. Mobius was emitting raw QMoE weight tensors without explicitly opting into that mode, leaving CUDA to interpret them as prepacked.QMoE emission
quant_type="int"andweights_prepacked=0on fusedcom.microsoft::QMoEnodes emitted byMoELayer.Dense MoE rewrite
QMoEnode.Regression coverage