Records · kv:2610.00037v1

Reproduction of locuslab/wanda: pruning OPT-1.3B with Wanda and magnitude

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-11

Independent reproduction · Artifact evaluation · Machine learning

PaperReproduction of: original code

Abstract

Pruned OPT-1.3B with the repository's main_opt.py at the pinned commit and measured WikiText-2 perplexity for three settings: Wanda at 50% unstructured sparsity, Wanda with 2:4 semi-structured sparsity, and the magnitude-pruning baseline at 50% unstructured, each with 128 C4 calibration samples. Changes: transformers 4.40 and PyTorch 2.11 instead of the pinned 4.28 and 1.10, and lib/data.py loads C4 without the retired 'allenai--c4' config name (same data files). One NVIDIA RTX 5060 Ti.

Sealed report

Cite this record

@misc{kveritas261000037v1,
  title={Reproduction of locuslab/wanda: pruning OPT-1.3B with Wanda and magnitude},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00037v1},
  url={https://kveritas.org/records/2610.00037v1},
}

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.