Transfer learning by only using MMARs

Hi, Is it possible to increase the classes of a pre-trained NGC model by e.g. changing the last layer (likely softmax) and use fine_tune.sh? I would like to “fine_tune.sh” the covid model - it’s not a binary task but a multi-class classification.
I know I can import the model into TF or Pytorch, edit the model, and export the model again - I was wondering if I can do it only using the MMAR’s?
Thanks!

Hi
Unfortunately, This kind of network surgery is not supported, yet. If you change the network you won’t be ale to load the weights and fine tune. you would be getting network errors.
Of course you can do that change and train from scratch.