Preprint
Aug 2026
QuaSAR: Quantization Compensation via Stable Activation-Aware Rank Truncation
This paper proposes a parameter-free truncated pseudoinverse solver which removes collapsed directions prior to inversion, and achieves 81.42\% top-1 accuracy, outperforming prior post-training methods and fine-tuning-based baselines.
Lin-Fa Lee, Yi-Yu Chang, Kuo-Hei Yeh
· 0 citations