Records · kv:2610.00011v1

Reproduction of KellerJordan/cifar10-airbench: airbench94_muon CIFAR-10 speedrun

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-07

Independent reproduction · Artifact evaluation · Machine learning

PaperReproduction of: original code

Abstract

Ran airbench94_muon.py unmodified at the pinned commit: one compile warmup run followed by the script's default 200 independent CIFAR-10 training runs (8 epochs each, Muon optimizer, torch.compile max-autotune), each evaluated with test-time augmentation; the script reports the mean and standard deviation of test accuracy over the 200 runs. 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{kveritas261000011v1,
  title={Reproduction of KellerJordan/cifar10-airbench: airbench94_muon CIFAR-10 speedrun},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00011v1},
  url={https://kveritas.org/records/2610.00011v1},
}

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.