Merged
Conversation
…model optimization Signed-off-by: Xin He <xin3.he@intel.com>
xin3he
changed the title
Contributor
Author
|
@copilot resolve the merge conflicts in this pull request |
# Conflicts: # auto_round/utils/missing_tensors.py Co-authored-by: xin3he <83260933+xin3he@users.noreply.github.com>
Contributor
Collaborator
|
/azp run Unit-Test-CUDA-AutoRound |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Contributor
There was a problem hiding this comment.
🟡 Changes recommended
The new NVFP4 dequantization path can crash on int8-packed weights due to a uint8-only unpack assertion, and the fused-expert split patterns/documented behavior are inconsistent for up_gate_proj.
Once you've addressed the issues Copilot identified, you can request another Copilot review.
Pull request overview
This PR refactors fused-expert tensor splitting and RTN weight packing utilities into auto_round.utils.model_free_utils, updates call sites/tests to the new import path, and extends the splitting/normalization logic to better handle special fused-expert naming and NVFP4 source tensors in model-free shard processing.
Changes:
- Moved
split_fused_expert_tensors/quantize_weight_rtnintoauto_round.utils.model_free_utilsand updated imports in core code + tests. - Enhanced fused-expert tensor splitting to recognize native FusedMoE suffix conventions and to support a “keep fused” passthrough for specific
model_typevalues. - Updated NVFP4 handling in model-free shard processing to dequantize NVFP4-packed sources when the target scheme is not NVFP4 (instead of letting packed companions fall through).
File summaries
| File | Description |
|---|---|
auto_round/utils/model_free_utils.py |
New home for fused-expert splitting + RTN packing; adds NVFP4 dequantization path and related plumbing. |
auto_round/utils/missing_tensors.py |
Removes local implementations and re-imports helpers from model_free_utils for backward compatibility. |
auto_round/utils/offload.py |
Updates runtime import to the new module for expert tensor splitting. |
auto_round/compressors/shard_writer.py |
Updates runtime import to the new module for fused expert expansion during shard writing. |
auto_round/utils/__init__.py |
Adds a wrapper-based re-export for copy_missing_tensors_from_source. |
test/unit/common/utils/test_missing_tensors.py |
Adds coverage for the new “keep fused for specific model_type” passthrough behavior. |
test/unit/test_cpu/export/test_qlinear_pack_clamp.py |
Updates test import path for quantize_weight_rtn. |
Review details
- Files reviewed: 7/7 changed files
- Comments generated: 3
- Review effort level: Lite
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
…ntiguity in dequantization Signed-off-by: Xin He <xin3.he@intel.com>
Collaborator
|
/azp run Unit-Test-CUDA-AutoRound |
|
Azure Pipelines successfully started running 1 pipeline(s). |
lkk12014402
approved these changes
Sep 9, 2026
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.
Description
This pull request refactors utility functions related to fused expert tensor splitting and weight quantization by moving them from
auto_round.utils.missing_tensorsto a new module,auto_round.utils.model_free_utils. This change improves code organization and modularity. The import paths are updated throughout the codebase to reflect this, and the moved functions are re-exported for backward compatibility.Refactoring and Code Organization:
split_fused_expert_tensorsandquantize_weight_rtnfromauto_round.utils.missing_tensorstoauto_round.utils.model_free_utils, and updated all internal imports and usages accordingly. [1] [2] [3] [4] [5]auto_round/utils/__init__.py, re-exportedcopy_missing_tensors_from_sourceas a wrapper to maintain backward compatibility with previous imports.Cleanup:
split_fused_expert_tensorsandquantize_weight_rtnfromauto_round.utils.missing_tensors.py. [1] [2]These changes do not affect functionality but improve maintainability and clarity by separating model-free utilities from tensor-missing logic.
Type of Change
Bug fix
Related Issues
Fixes or relates to #
Checklist Before Submitting
/azp run Unit-Test-CUDA-AutoRound.