onnxruntime ignores explicit local-function overload

This repository contains a minimal .onnx proof of concept showing that onnxruntime ignores the explicit overload field when resolving model-local functions.

Files

explicit_overload_ignored.onnx: minimal PoC
generate_poc.py: regenerates the PoC
reproduce.py: runs the model in onnxruntime

Reproduction

python generate_poc.py
python reproduce.py

Expected highlights

graph calls local::F with overload="ov1"
runtime output is [-2.], meaning ov2 was executed instead