marcoabreu commented on issue #11626: [MXNET-651] MXNet Model Backwards 
Compatibility Checker
   Please note that we are interested in the model architecture itself. Thus,
   I'd like to propose that we only use random weights and run inference on
   randon data - I don't know if we actually have to train at all. Then save
   that model as well as the input and output. Then run forward on the latest
   version and ensure the output is identical.
   We don't care whether the values actually make sense. We are rather
   interested in whether they are identical. This means that we don't need the
   overhead of real models and training. I'm happy to get corrected here if my
   assumption is wrong.

This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:

With regards,
Apache Git Services

Reply via email to