tuanle618/eqgat-diff ? reverse-engineered prompt

Build me the code for this research project about equivariant diffusion based 3D molecule generation.

I want a Python codebase that can run unconditional sampling, conditional sampling for properties like polarizability, and conditional sampling in a protein pocket setting. It should be able to generate 3D molecular structures over a diffusion style trajectory, and save outputs that can be turned into the kind of animated sampling examples shown in the README.

Please make it practical to run from the command line with clear entry points for training, sampling, and inference. Include sensible defaults, basic config handling, and a way to load and use pretrained weights if they exist. If any details are unclear, look up the current docs or related papers online and make a reasonable implementation.

Keep the project organized so the main code lives in the package directory, and make sure the sampling scripts can work with the provided data and weights folders without extra manual setup.