Records · kv:2610.00004v1

Reproduction of kuangliu/pytorch-cifar: ResNet18 on CIFAR-10

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-07

Independent reproduction · Benchmark or leaderboard entry · Computer vision

Reproduction of: original code

Abstract

Trained ResNet18 on CIFAR-10 with the repo's main.py (SGD lr 0.1, momentum 0.9, weight decay 5e-4, batch 128, random crop and flip, cosine schedule) for 50 epochs instead of 200, with the cosine T_max set to match. Two harness-only edits: ResNet18 selected in place of the default SimpleDLA, and the terminal progress bar prints only its final line per pass when stdout is not a TTY. Test accuracy evaluated every epoch. CIFAR-10 downloaded by torchvision. Single NVIDIA RTX 5060 Ti 16 GB, PyTorch 2.11 cu128, Python 3.12.

Sealed report

Cite this record

@misc{kveritas261000004v1,
  title={Reproduction of kuangliu/pytorch-cifar: ResNet18 on CIFAR-10},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00004v1},
  url={https://kveritas.org/records/2610.00004v1},
}

References

1. [1] Mamadou K. Keita and Christopher Homan. Computer Science Conferences Should Require Nonrepudiable Experimental Results. NeurIPS 2026 Position Paper Track. arXiv:2605.08586.

Sealed with K-Veritas [1]. Listed, not endorsed. A record shows these results came from the sealed code, unchanged. Whether the work is sound is for the reader to judge.