Records · kv:2610.00031v1

Reproduction of microsoft/BitNet: BitNet b1.58 2B CPU inference throughput

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-10

Independent reproduction · Artifact evaluation · Natural language processing

PaperReproduction of: original code

Abstract

Built bitnet.cpp at the pinned commit with setup_env.py (clang 18, i2_s kernels) and ran the README benchmark utils/e2e_benchmark.py on the official BitNet-b1.58-2B-4T GGUF: 512-token prompt processing and 128-token generation, batch 1, 8 threads, CPU only, 5 repetitions. At this commit e2e_benchmark.py exits with status 1 even on success, so success was judged by its result table. Intel Xeon W-2145.

Sealed report

Cite this record

@misc{kveritas261000031v1,
  title={Reproduction of microsoft/BitNet: BitNet b1.58 2B CPU inference throughput},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00031v1},
  url={https://kveritas.org/records/2610.00031v1},
}

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.