UniNAS is a highly modular PyTorch framework with a focus on Neural Architecture Search (NAS), developed by Kevin Laube.
It's main focus is the consideration of everything as a hyper-parameter, including the optimizer choice, schedules, regularization, and even network topologies. Dynamically generated argument parsing trees solve the issue of arbitrarily complex and prior unknown configurations, making it even possible to run powerful tasks without writing any code.
An introductionary presentation is available here
The code is open source on GitHub