Records · kv:2610.00007v1

Reproduction of sthalles/SimCLR: SimCLR ResNet-18 on CIFAR-10 with linear evaluation

K-Veritas Team

Submitted by K-Veritas Team

Published 2026-10-07

Independent reproduction · Artifact evaluation · Computer vision

PaperReproduction of: original code

Abstract

Pretrained a ResNet-18 SimCLR model on CIFAR-10 with run.py (50 epochs, batch 256, Adam lr 3e-4, temperature 0.07, fp16 mixed precision, the script's fixed seed 0), then ran the repo's linear evaluation from feature_eval/mini_batch_logistic_regression_evaluator.ipynb as a script: frozen backbone, logistic-regression head trained with Adam for 100 epochs on CIFAR-10, top-1/top-5 test accuracy each epoch. The notebook code is unchanged except that it loads the checkpoint just trained instead of downloading a published one; run.py gained one print per epoch of the values it already logs. 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{kveritas261000007v1,
  title={Reproduction of sthalles/SimCLR: SimCLR ResNet-18 on CIFAR-10 with linear evaluation},
  author={K-Veritas Team},
  year={2026},
  howpublished={K-Veritas Records},
  note={kv:2610.00007v1},
  url={https://kveritas.org/records/2610.00007v1},
}

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.