Hello, everyone, I have been working on introducing tests to check model compatibility and sanity across different MXNet versions. The goal of the project was to ensure that models trained on earlier versions of MXNet are backward compatible with the latest release candidate version: they can be loaded by latest RC and also are consistent with regards to the inference outputs.
I have designed and implemented first version of Model Backwards Compatibility Checker and you can find more details about it in this cWiki document <https://cwiki.apache.org/confluence/display/MXNET/Model+Backwards+Compatibility+Check> . I would like to invite more ideas/suggestions from the community on how can we further improve and extend this project. Best, Piyush
