ViniciusMikuni/OmniLearned — reverse-engineered prompt

Build me a Python package and command line tool for the OmniLearned project so I can reproduce the paper results and also try it on my own jet physics data.

I want it to be easy to install, with commands to download supported public datasets, including a big pretrain option with a clear storage warning, then train either a classifier or a generative model, evaluate the results, and save useful outputs like predictions and labels. Please include a simple way to load and use a pre trained checkpoint, plus an option to fine tune on a custom dataset.

It should feel practical for researchers, with sensible defaults, examples for small training runs, and helper scripts for single GPU and multi GPU or SLURM style runs. Add enough docs so someone can follow the full flow from data download to training to evaluation without digging through the code, and include tests for the main commands. If anything is unclear, look up the current docs online and keep the interface consistent.