English

Robust Ultra Low-Bit Post-Training Quantization via Stable Diagonal Curvature Estimate

Machine Learning 2026-04-16 v1

Abstract

Large Language Models (LLMs) are widely used across many domains, but their scale makes deployment challenging. Post-Training Quantization (PTQ) reduces memory footprint without retraining by leveraging a small calibration set. Recent Hessian-based PTQ methods compensate quantization error via cross-channel dependencies, but such approaches degrade at low bit-widths due to noisy curvature estimates from limited calibration data. We propose DASH-Q, a robust PTQ framework using diagonal Hessian approximation and iterative weighted least squares. By discarding noise-prone dependencies, DASH-Q filters sampling noise while prioritizing the preservation of salient feature power. We outperform other PTQ baselines in ultra low-bit regime, improving zero-shot accuracy by 7.01% on average and up to 14.01% over the strongest baselines across five baseline LLM models, while showing robust and stable performance with very small calibration data.

Keywords

Cite

@article{arxiv.2604.13806,
  title  = {Robust Ultra Low-Bit Post-Training Quantization via Stable Diagonal Curvature Estimate},
  author = {Jaemin Kim and Sungkyun Kim and Junyeol Lee and Jiwon Seo},
  journal= {arXiv preprint arXiv:2604.13806},
  year   = {2026}
}

Comments

EUROMLSYS 2026

R2 v1 2026-07-01T12:10:39.468Z