Justin Chu

I added a todo in the PR for tensor_tensor. We don't want to do manual type promotion to complicate the implementation unless absolutely necessary (will be fixed with a more...

### Summary ONNX Runtime raises `` when executing test `ops_test.TestOutputConsistencyFullGraphCPU.test_output_match_opinfo__concat_cpu_float32` in ONNX Script `TorchLib`. To recreate this report, use ```bash CREATE_REPRODUCTION_REPORT=1 python -m pytest onnxscript/tests/function_libs/torch_lib/ops_test.py -k test_output_match_opinfo__concat_cpu_float32 ``` ### To...

cc @gramalingam

> @justinchuby > > Perhaps to pick up the discussion here again, Python 3.8 is now end of live. What does this mean for us? Should we discuss this in...

Thanks. I personally don't have a concern. I would invite more people to chime in.

1. It would be nice if we can follow https://scientific-python.org/specs/spec-0000/. My hesitation is that ONNX is an important dependency for PyTorch which supports a wider range of Python versions. To...

The torch.onnx component is dependent on onnx so onnx is needed when users what to export models from PyTorch. We need the latest version of onnx to have access to...

Maybe just bump protobuf version ?

This just looks like a compiler version update. I think we should maintain the ability to compile with the latest os

SGTM!