mzjb/deeph-pack ? reverse-engineered prompt

Build me a Python package for predicting density functional theory Hamiltonians with a deep learning model, like the DeepH project described in the paper. It should let a user prepare datasets from supported DFT outputs, train a model, and run predictions on new materials systems. Please make it work with the kinds of inputs mentioned in the README, especially ABACUS, OpenMX, FHI aims, and SIESTA, and include a simple example workflow so someone can try it end to end.

I’d also like the repo to feel usable, with clear install steps, basic command line or script entry points, and documentation that explains how to set up the environment and run the demo. If anything about the current implementation is unclear, look up the current docs online if you need to, and keep the code organized so it’s easy to extend later.