Hi I am trying to use hydra multirun on top of Nvidia Modulus, I want to store the outputs in a folder name of my choice. What changes do I need to make? I think most of the paths are defined in the solver class. https://gitlab.com/nvidia/modulus/modulus/-/blob/release_22.09/modulus/solver/solver.py
This should be adjustable using Hydra. Hydra conf changes the programs working location on init, so its controlling where the output folders are. The output folder is defined under the config group /hydra/run/dir I believe (for a single run) and /hydra/sweep/dir (for multirun). You could do this using command line, overrides the config or in the source code here.
If you’re looking at the Hydra docs, just remember the docker file is currently using a slightly older version of Hydra that the current release. Most API is consistent however.
So, my code was written in Modulus 22.03. That particular version works with Hydra 1.1. I managed to scrap hydra using the Jupyter notebook tutorial. Then I created a wrapper of Hydra-zen. This package is useful when the hyperparameters are changing (using an algorithm) as we proceed with the multirun.
1 Like
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.