[BUG] The pass "eliminate_nop_expand" causes output mismatch
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 834
- Forks
- 109
- Avg merge
- 6h 55m
- Merged PRs (30d)
- 2
Description
[BUG] The pass "eliminate_nop_expand" causes output mismatch
Issue
Running the single pass eliminate_nop_expand with onnxoptimizer 0.4.2 changes numerical outputs.
Environment
- Ubuntu 20.04
- Python 3.10
- onnx==1.19.0
- onnxruntime==1.23.2
- onnxoptimizer==0.4.2(latest)
Repro steps (run from this folder)
- Create a Python environment (Python 3.10) and install dependencies:
python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -r requirements.txt
- Optimize a case with only
eliminate_nop_expand(writesmodel.opt.onnxnext tomodel.onnx):
python optimize_model.py --case ./case_08971_seed91292368
- Differential test original and optimized models with same oracle inputs (ORT graph optimization disabled):
python diff_test.py --case ./case_08971_seed91292368
Expected
eliminate_nop_expand should be semantics-preserving. Applying this pass alone should not change any output values.
Differential Test Output Details
- case_08971_seed91292368 — output[1] max_abs=6.000e+00 max_rel=6.000e+08; output[5] max_abs=1.000e+00 max_rel=1.000e+00; overall mismatch.
Case: case_08971_seed91292368
output[0]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(1, 55, 1, 1, 1)
output[1]: max_abs=6.000e+00, max_rel=6.000e+08, shape=(2, 55, 2, 1, 2)
output[2]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(2, 55, 2, 1, 2)
output[3]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(41, 55, 2, 1, 2)
output[4]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(41, 55, 2, 1, 6)
output[5]: max_abs=1.000e+00, max_rel=1.000e+00, shape=(41, 55, 2, 1, 6)
output[6]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(40590,)
output[7]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(1,)
output[8]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(1, 1)
output[9]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(1,)
output[10]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(1, 1)
output[11]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(1,)
output[12]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(41, 55, 18, 1)
output[13]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(41, 55, 18, 1)
output[14]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(41, 55, 1, 1)
output[15]: max_abs=0.000e+00, max_rel=0.000e+00, shape=(41, 55, 1, 1)
Overall: max_abs=6.000e+00, max_rel=6.000e+08
Attachments
requirements.txtoptimize_model.pydiff_test.py(runs original and optimized with same oracle inputs; ORT graph optimizations disabled)run_optimized_model.py(executes optimized model with oracle inputs)case_08971_seed91292368/(original model and oracle inputs)
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by running optimize_model.py and diff_test.py for case_08971_seed91292368 with the listed dependencies, then inspect the eliminate_nop_expand pass and the attached model. Compare the original and optimized outputs, especially output[1] and output[5]. Done means the pass preserves all reported output values.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp, python
- Domain
- performance, tooling
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 42/100