Pre-trained and fine-tuned checkpoints for
MapPFN: Learning Causal Perturbation Maps in Context (Sextro et al., 2026).
All checkpoints share the same MMDiT architecture (~25M parameters) and differ only in training data. See the
GitHub repository for inference and fine-tuning code.
1@article{sextro2026mappfn,
2 title = {{MapPFN}: Learning Causal Perturbation Maps in Context},
3 author = {Sextro, Marvin and K\l{}os, Weronika and Dernbach, Gabriel},
4 journal = {arXiv preprint arXiv:2601.21092},
5 year = {2026}
6}
