@article{gsqhighlyaccuratelowprecisionscalarquant, title = {GSQ: Highly-Accurate Low-Precision Scalar Quantization for LLMs via Gumbel-Softmax Sampling}, author = {Alireza Dadgarnia and Soroush Tabesh and Mahdi Nikdan and Michael Helcig and Eldar Kurtic and Maximilian Kleinegger and Dan Alistarh}, year = {2026}, eprint = {2604.18556}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2604.18556}, }