ONNX Runtime will be releasing at the end of August and there are a few important bug fixes that would be ideal to make their way into an ONNX 1.14.1 bug-fix release.

@gramalingam and I called these out on the Weekly Sync this morning, but I wanted to open an issue for tracking and discussion, and to identify other potential fixes to take into the release.

Necessary for ONNX Runtime:

Fixes regarding a number of shape inferencing bugs (these were surfaced by ONNX Script and/or the TorchDynamo-based ONNX exporter):

• Fix shape data propagation function to handle missing optional parameters #5219
• Fix a couple of shape inference issues #5223
• Extend function type inference to handle missing optional parameters #5169

Other PRs nice-to-have addressing misc bugs:

• Fix check_tensor to work with large models on Windows #5227
• Fix check_tensor to work with large models on UNIX #5286

@cc @yuanyao-nv @gramalingam @jcwchen @liqunfu @BowenBao @justinchuby

For reference, these are all PRs merged since 1.14.0.