Data for "Generative artificial intelligence for reconstructing neutron-star matter"

Trained diffusion models, training and validation data, and posterior samples of the neutron-star equation of state from "Generative artificial intelligence for reconstructing neutron-star matter" (J. Yu. Panteleeva, H. Alharazin, E. Epelbaum, arXiv:2608.17457). Each sample is the squared speed of sound c_s²(n_B) on 200 points in n_B/n_0 ∈ [0.5, 8], drawn from a denoising-diffusion prior with chiral-EFT anchors and reweighted by the pQCD, NICER, GW170817 and maximum-mass likelihoods. Contents: trained_models.zip (baseline model M0 and robustness variants MI–MIII, with training logs); training_and_validation_data.zip (c_s² curves of classes 1–13); posteriors.zip (baseline and six robustness posteriors, with importance weights, individual log-likelihood terms and mass–radius–tidal-deformability curves); README.md. Code and reproduction instructions: https://github.com/JuliaYu24/Neutron_Star_Matter_Diffusion. The posteriors depend on published inputs from other groups; see README_EXTERNAL_DATA.md in the code repository for the sources to cite.