RuRo edited a comment on pull request #18525: URL: https://github.com/apache/incubator-mxnet/pull/18525#issuecomment-646597349
@eric-haibin-lin can you explain, why did you remove the conversions for `Add`, `Sub`, `Mul` and `Div` operators from onnx to mxnet? Just because it uses the Module API? Isn't this a bit of an overkill? :confused: AFAIK, the module API is used here just to infer the shapes of inputs, so it should be probably possible to replace the `get_input_shape` call with some alternative function, which doesn't use the Module API. (Probably, something along the lines of [MXNetGraph.get_outputs](https://github.com/apache/incubator-mxnet/blob/cb54a4a99463b23b8abaa2629661954c4ba3c60b/python/mxnet/contrib/onnx/mx2onnx/export_onnx.py#L119)) P.S. I just realized that you also completely removed the `test_models` ONNX tests which were also using the Model API. I think, it was previously mentioned that support for ONNX will **not** be dropped in MXNet 2.0, so please don't just delete things. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
