Records · kv:2610.00002v1

Reproduction of omihub777/ViT-CIFAR: ViT from scratch on CIFAR-10

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-07

Independent reproduction · Artifact evaluation · Computer vision

PaperReproduction of: original code

Abstract

Trained the repo's 6.3M-parameter Vision Transformer from scratch on CIFAR-10 with the README command (200 epochs, Adam lr 1e-3, 5 warmup epochs then cosine, label smoothing, AutoAugment, 16-bit mixed precision, seed 42), logging with the CSV logger since no Comet key is used. Compatibility changes for current Python and PyTorch: pytorch-lightning 1.5.10 instead of the pinned 1.2.1, np.int replaced by int in autoaugment.py, and the per-epoch lr log call given on_epoch=True. Validation accuracy each epoch from the logged metrics. 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{kveritas261000002v1,
  title={Reproduction of omihub777/ViT-CIFAR: ViT from scratch on CIFAR-10},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00002v1},
  url={https://kveritas.org/records/2610.00002v1},
}

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.