Muhtasim-Munif-Fahim/Green-NAS ? reverse-engineered prompt

Build me a Python project for edge weather forecasting that can run a multi objective neural architecture search and compare the best models against a few hand made baselines.

I want the app to download hourly weather data for a bunch of cities from Open Meteo, clean it up, train on source cities, then evaluate on target cities with transfer learning. It should search for compact models that balance accuracy and size, and save the best results and plots so I can see the Pareto front, training curves, and feature importance.

Please include scripts to download and preprocess the data, run the NAS search, train the baseline models, train the discovered models, run transfer learning experiments, and verify efficiency like latency, model size, and FLOPs. Keep it easy to run from the command line, with a clear config file for cities and settings.

If anything is unclear, look up current docs online if you need to, and make the code organized so it is easy to extend later.