wyky481l/MMCD — reverse-engineered prompt

I want this repo turned into a working training setup for the MMCD therapeutic peptide generator from the README. Please make it easy for me to run the AMP example end to end with the sequence fasta files and the peptide structure pdb files, using the multimodal idea of sequence plus structure. If anything is missing, wire it up so the data goes in the expected folders, the training script runs cleanly, checkpoints are saved to an output folder, and I can tell when training is actually progressing.

Please also add a simple way to evaluate or inspect the generated peptides after training, even if it is just basic outputs and metrics already supported here. I am not trying to redesign the model, I just want the project to be runnable and understandable by someone who is not deep in this codebase. If you need to, check the paper or current docs online so the setup matches the intended MMCD workflow.