Records · kv:2610.00009v1

Reproduction of pytorch/examples: Transformer word-level language model on WikiText-2

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

Trained the word_language_model example from the pinned commit unmodified: a Transformer language model (default size: 200-dim embeddings, 2 layers, 2 heads, dropout 0.2) on the WikiText-2 copy shipped in the repo, 20 epochs, initial learning rate 5 with the script annealing schedule, seed 1111, on the GPU (--accel, which replaces the older --cuda flag). Validation loss and perplexity per epoch and the final test loss and perplexity are parsed from the script output. Single NVIDIA RTX 5060 Ti 16 GB, PyTorch 2.11 cu128, Python 3.12.

Sealed report

Cite this record

@misc{kveritas261000009v1,
  title={Reproduction of pytorch/examples: Transformer word-level language model on WikiText-2},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00009v1},
  url={https://kveritas.org/records/2610.00009v1},
}

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.