---
license: mit
pipeline_tag: image-feature-extraction
---

# Spherical Leech Quantization for Visual Tokenization and Generation

[![arXiv](https://img.shields.io/badge/arXiv%20paper-2512.14697-b31b1b.svg)](https://arxiv.org/abs/2512.14697) 
[![Project Page](https://img.shields.io/badge/Project%20Page-Website-lightblue.svg)](https://cs.stanford.edu/~yzz/npq/) 
[![code](https://img.shields.io/badge/github-zhaoyue%2D-zephyrus/InfinityCC-blue?logo=github)](https://github.com/zhaoyue-zephyrus/InfinityCC) 
<!-- The following GitHub repository for 'bsqvit' seems less directly tied to the primary model checkpoint in this repository compared to 'InfinityCC', so it is commented out for clarity.
[![code](https://img.shields.io/badge/github-zhaoyue%2D-zephyrus/bsqvit-blue?logo=github)](https://github.com/zhaoyue-zephyrus/bsq-vit)&nbsp;
-->

This model implements **Spherical Leech Quantization ($\Lambda_{24}$-SQ)**, a novel non-parametric quantization method for visual tokenization and generation. Leveraging the high symmetry and even distribution of the Leech lattice, $\Lambda_{24}$-SQ simplifies training and improves reconstruction-compression trade-offs. It outperforms prior art (like BSQ) in image tokenization and compression tasks and extends its benefits to state-of-the-art autoregressive image generation frameworks.