Records · kv:2610.00010v1

Reproduction of karpathy/minbpe: BPE tokenizer training and tests

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-07

Independent reproduction · Benchmark or leaderboard entry · Natural language processing

Reproduction of: original code

Abstract

Ran train.py unmodified at the pinned commit: trains the Basic and Regex byte-pair-encoding tokenizers to a 512-token vocabulary on the repository's sample text. The run then loads both saved tokenizers, checks that encoding round-trips the text, measures the compression ratio (bytes per token), and runs the repository's test suite, which compares against tiktoken. CPU only, one machine.

Sealed report

Cite this record

@misc{kveritas261000010v1,
  title={Reproduction of karpathy/minbpe: BPE tokenizer training and tests},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00010v1},
  url={https://kveritas.org/records/2610.00010v1},
}

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.