According to Conv op in ONNX spec, kernel_shape is an attribute to show the shape of the convolution kernel: https://github.com/onnx/onnx/blob/main/docs/Operators.md#Conv However, it's skipped in torchlib implementation, resulting in absent of kernel_shape attribute in onnx model. op.Conv in https://github.com/microsoft/onnxscript/blob/main/onnxscript/function_libs/torch_lib/ops/core.py Reference implementation of legacy symbolic function: https://github.com/pytorch/pytorch/blob/4fbc0c586f979ab3b27737b76cf4a01c1e095017/torch/onnx/_internal/torchscript_exporter/symbolic_opset9.py#L2399
According to Conv op in ONNX spec, kernel_shape is an attribute to show the shape of the convolution kernel: https://github.com/onnx/onnx/blob/main/docs/Operators.md#Conv
However, it's skipped in torchlib implementation, resulting in absent of kernel_shape attribute in onnx model.
op.Conv in https://github.com/microsoft/onnxscript/blob/main/onnxscript/function_libs/torch_lib/ops/core.py
Reference implementation of legacy symbolic function: https://github.com/pytorch/pytorch/blob/4fbc0c586f979ab3b27737b76cf4a01c1e095017/torch/onnx/_internal/torchscript_exporter/symbolic_opset9.py#L2399