Lightweight wrappers for training, sampling, and evaluation.
train_cosmodiff.py: callscosmo_diffusion/scripts/cosmodiff_train.py.sample_cosmodiff.py: loads a checkpoint and writes generated samples to.npy.evaluate_samples.py: computes histogram, power-spectrum, and nearest-neighbor metrics.reproducibility_eval.py: compares multiple generated sample sets.plot_reproducibility_figure.py: makes a Figure-1-style pairwise reproducibility plot from a manifest of generated sample files.plot_generalizability_sscd.py: computes an SSCD near-copy/generalization curve for one architecture across training-set sizes.slurm/train_template.sbatch: sanitized Slurm template. Copy it into an ignored local path and edit account, partition, paths, and walltime there.