I am trying to use Nvidia Modulus v2209 and to find optimized hyper-parameters in my neural network architecture (i.e. number of layers or hidden neurons).
I wish to ask how to use AutoML in Nvidia Modulus and find the optimized model with minimum validation error.
Such as: ‘l2_relative_error_u’
Unfortunately I don’t think we have tried AutoML from the Modulus teams side, not sure if anyone in the community has either. Perhaps a primitive alternative is Hydra config multi-run which allows you to do easy hyper-parameter sweeps of different parameters controlled in your YAML file. An example in our docs.
Dear @ngeneva , thank you for your notice, will try that!