English

Technical Report: Activation Residual Hessian Quantization (ARHQ) for Low-Bit LLM Quantization

Machine Learning 2026-05-04 v1 Computation and Language Computer Vision and Pattern Recognition

Abstract

We present Activation Residual Hessian Quantization (ARHQ), a post-training weight splitting method designed to mitigate error propagation in low-bit activation-weight quantization. By constructing an input-side residual Hessian from activation quantization residuals (G_x), ARHQ analytically identifies and isolates error-sensitive weight directions into a high-precision low-rank branch. This is achieved via a closed-form truncated SVD on the scaled weight matrix W G^{1/2}_x . Experimental results on Qwen3-4B-Thinking-2507 demonstrate that ARHQ significantly improves layer-wise SNR and preserves downstream reasoning performance on ZebraLogic even under aggressive quantization. The code is available at https://github.com/BeautMoonQ/ARHQ.

Cite

@article{arxiv.2605.00140,
  title  = {Technical Report: Activation Residual Hessian Quantization (ARHQ) for Low-Bit LLM Quantization},
  author = {YiFeng Wang and Zhun Sun and Keisuke Sakaguchi},
  journal= {arXiv preprint arXiv:2605.00140},
  year   = {2026}
}
R2 v1 2026-07-01T12:44:23.566Z