G-STAR: Group-Scaled Token Capacity for Deeply Compressed Autoregressive Generation
Paper · https://arxiv.org/abs/2611.13004
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.
First seen by Oriel on .
AI summary
This paper proposes GSTAR, a spatial–group co-scaling framework and semi-autoregressive decoder that enable highly compressed discrete representations for fast, high-quality autoregressive image generation.
- Under a fixed nominal information budget, the proposed principle scales product-quantization group size quadratically with per-axis spatial compression.
- GSTAR supports 32× and effective 64× compression while preserving reconstruction, understanding, and generation performance.
- Its semi-autoregressive decoder combines parallel feature computation with lightweight causal modeling to address intra-group dependencies.
- At 1024×1024 resolution, the model generates an image in 2.5 seconds on one NVIDIA H200, claiming a 20× speedup over prior discrete autoregressive models.
- The models perform competitively across GenEval, DPG-Bench, PRISM, LongText-Bench, and Qwen-Image-Bench.
#autoregressive generation #image generation #product quantization #model compression
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hugging Face Papers: 46, latest observation ; +6 over observed period since
Comments
Comment count reported by the source.
- Hugging Face Papers: 3, latest observation ; +2 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| arXiv AI Papers | oai:arXiv.org:2611.13004v1 | https://arxiv.org/abs/2611.13004 Read discussion |
| Hugging Face Papers | 2611.13004 | https://arxiv.org/abs/2611.13004 Read discussion |