Justin Chu

I wonder if the constraints are formally expressed and if verification code can be generated.

fp8 related: https://github.com/openxla/xla/discussions/22

- https://github.com/onnx/onnx/issues/4287

I don’t think we have an option yet. https://github.com/onnx/onnx/blob/main/CMakeLists.txt is what we will have to edit. @jcwchen is that right? Contribution/help would be much appreciated!

This is a previously failed attempt: https://github.com/onnx/onnx/pull/5053

I second this, for the reasons above

@liqunfu is this the same as the If operator error?

I think youtube is good in general. For access in China we can consider a Chinese service like bilibili.

The model text ```cpp < ir_version: 8, opset_import: ["pkg.onnxscript.torch_lib" : 1, "" : 18, "pkg.onnxscript.torch_lib.common" : 1], producer_name: "pytorch", producer_version: "2.2.0" > main_graph (float16 input_0) => (float16 _val_3) { _val_1...

Essentially the equation `...ik, ...j -> ij`