Modularize Model Configs

[rhubarbwu]

I sent the following email to Danial at Stratum AI.

Hi Danial,

It's been a while since we spoke, but I'm still reflecting on how much I learned from you. We recently presented our [ENAS work](https://github.com/rusbridger/ENAS-Experiments) at AAAI and have begun new extensions that require better experiment modularization. Could you remind me of the config/hyperparameter management tool you suggested instead my numerous python files? We'll probably continue using PyTorch, but would it work for other frameworks like TF and JAX?

One consideration is how I organize experiments.
The configs can all live in main repo itself, but then it'd have to track all the experiments on top of each other in a longer revision tree.
Alternatively, I'm thinking of storing these configs in GitHub Gists that can be individually imported independently.
What do you think?

Rupert