This repository contains a 2-bit quantized version of
meta-llama/Llama-2-7b-hf using the
BPDQ (Bit-Plane Decomposition Quantization) algorithm.
BPDQ is a post-training quantization (PTQ) method that constructs a variable quantization grid via bit-plane decomposition and scalar coefficients. By iteratively refining these using second-order information, it maintains high fidelity even in extremely low-bit regimes (2–3 bits) where conventional PTQ methods typically degrade.
1@article{chen2026bpdq,
2 title={BPDQ: Bit-Plane Decomposition Quantization on a Variable Grid for Large Language Models},
3 author={Chen, Junyu and Li, Jungang and Xiong, Jing and Wang, Wenjie and Yang, Qingyao and Xiao, He and Li, Zhen and Wu, Taiqiang and Chen, Mengzhao and Peng, Zhen and others},
4 journal={arXiv preprint arXiv:2602.04163},
5 year={2026}
6}
