| title | G-STAR: Group-Scaled Token Capacity for Deeply Compressed Autoregressive Generation |
|---|---|
| authors | Xu Huang, Ye Huang, Xiaotong Li, Zhengrong Yue, Jialv Zou, Minhao Jing, Haodong Li, Wenhao Yan, Yaxin Luo, Duomin Wang, Daquan Zhou |
| date | 2026-11-08 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2611.13004 |
| affiliation | LongCat |
| upvotes | 45 (Hugging Face Daily Papers, 2026-11-15) |
| links | abs pdf hf code |
| cite | |
| related | WBench: A Comprehensive Multi-turn Benchmark for Interactive Video World Model Evaluation Beyond Final Scores: A Systematic Evaluation of Agents for Long-Horizon AI Research and Development LongCat-DeepResearch Technical Report CAST: Game Solvers as Turn-Level Teachers for LLM Agents Data World: Scaling Data Environments from Scratch |
| abstract | Discrete autoregressive models offer a scalable framework for image generation. However, they face a fundamental trade-off: long token sequences preserve representation quality but slow generation, while aggressive spatial compression reduces representation capacity. Based on product quantization, we show that, under a fixed nominal information budget, group size should scale quadratically with the per-axis spatial compression factor. Building on this principle, we propose GSTAR, a spatial--group co-scaling framework that enables \(32\times\) and even effective \(64\times\) compression while preserving strong reconstruction, understanding, and generation performance. Since larger groups introduce intra-group dependencies that must be modeled efficiently, we analyze their impact on inference latency and generation quality and introduce a semi-autoregressive decoder that combines parallel feature computation with lightweight causal modeling. With these designs, our highly compressed models perform competitively with leading models across five representative text-to-image benchmarks (GenEval, DPG-Bench, PRISM, LongText-Bench, and Qwen-Image-Bench). At \(1024\times1024\) resolution, our model delivers a \(20\times\) speedup over prior discrete autoregressive models, generating each image in 2.5 seconds on a single NVIDIA H200. Together, these results show that highly compressed discrete representations can enable high-quality autoregressive generation at low latency. |
bibtex
@misc{huang2026g,
title={G-STAR: Group-Scaled Token Capacity for Deeply Compressed Autoregressive Generation},
author={Xu Huang and Ye Huang and Xiaotong Li and Zhengrong Yue and Jialv Zou and Minhao Jing and Haodong Li and Wenhao Yan and Yaxin Luo and Duomin Wang and Daquan Zhou},
year={2026},
eprint={2611.13004},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2611.13004}
}