DGLD4Energetic: code, data, and model checkpoints for Domain-Gated Latent Diffusion (energetic-materials discovery)

DGLD4Energetic - complete reproducibility package for Domain-Gated Latent Diffusion: Generative Inverse Design of HMX-Class Energetic Materials with First-Principles Validation (Y. Aperstein & A. Apartsin). This version bundles the full pipeline alongside the trained checkpoints: Model checkpoints (inherited from the first version): LIMO baseline, domain-gated latent diffusion generators (v3, v4b), score models (v3e, v3f), tokenizer vocabulary and metadata. Code - DGLD4Energetic-code.zip: runnable training and generation code, documentation, license, and citation (Apache-2.0). Results and provenance - DGLD4Energetic-data.zip: per-experiment result files, sweep logs, and provenance for the figures and tables in the paper (CC-BY-4.0). Raw training datasets (gzipped CSV): the 65,980-row labelled master, the ~700k-row unlabelled corpus, and the 1.22M-row motif-augmented set. Hard negatives - hard_negatives_v3e.pt: the v3e labelled latent tensor carrying the 918 mined hard-negative examples used for domain gating. Code is licensed Apache-2.0; data and checkpoints are CC-BY-4.0. Source and issue tracker: github.com/ApartsinProjects/DGLD4Energetic.