HEP4M tokenised COCOA single-jet dataset and model checkpoints

DOI: 10.5281/zenodo.22917591. Code: github.com/FM-for-HEP/hep-any2any. Licence: data and model checkpoints CC-BY-4.0; code (hep4m_data.py and the code repository) Apache-2.0. Here we provide the tokenised single-jet dataset used to train and evaluate the HEP4M and nanoHEP models in "Prompting Particle Physics: Tokenized Multi-modal Foundation Models for Combinatorially Many Tasks" (Kakati, Murnane et al., 2026; arXiv:2609.31862). It also holds the seven tokenisers, the raw ROOT files and HGPflow predictions for the evaluation events, and the seven model checkpoints behind the paper's tables. Every event is a single jet simulated with COCOA. Each event is stored in seven modalities: tracks, topoclusters, truth particles, HGPflow particles, the truth jet, calorimeter cells, and the per-cell truth charged-energy fraction. Every object is a short row of integer tokens produced by a frozen per-modality tokeniser. The full data card, including the file list, modalities, tokeniser provenance and loading instructions, is in README.md in this record.