Research map / Image generation / Autoregressive and latent generation
Parallel Jacobi Decoding for Fast Autoregressive Image Generation
Boya Liao, Ying Li, Siyong Jian, Huan Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Parallel Jacobi Decoding (c=11) instead of SJD | Lumina-mGPT 7B · MS-COCO | step acceleration: 2.23× -> 6.35× | Improved |
| use PJD (c=6) instead of GSD | LlamaGen-XL 7B · MS-COCO | step acceleration: 2.67× -> 4.81× | Improved |
| use PJD (c=16) instead of GSD | Lumina-mGPT 7B · MS-COCO | FID: 33.41 -> 31.94 | Improved |
| add Row-Causal Mask (ablation) | PJD (c=11) · MS-COCO, Lumina-mGPT | step acceleration: 6.07× -> 6.35× | Improved |
| add Row-Causal Mask (ablation) | PJD (c=11) · MS-COCO, Lumina-mGPT | FID: 32.65 -> 32.38 | Improved |
| add Row-Causal Mask (ablation) | PJD (c=6) · MS-COCO, LlamaGen-XL | step acceleration: 4.41× -> 4.81× | Improved |
| add Row-Causal Mask (ablation) | PJD (c=6) · MS-COCO, LlamaGen-XL | FID: 45.67 -> 45.12 | Improved |
| increase context token count c from 9 to 16 (ablation) | PJD · MS-COCO, Lumina-mGPT | FID: 32.98 -> 31.94 | Improved |
| increase context token count c from 9 to 16 (ablation) | PJD · MS-COCO, Lumina-mGPT | step acceleration: 7.18× -> 4.95× | Hurt |
| vary top-k sampling (ablation) | PJD · MS-COCO, Lumina-mGPT | step acceleration | No clear effect |
| vary CFG scale (higher) (ablation) | PJD · MS-COCO, Lumina-mGPT | FID | Improved |
| increase image resolution to 1024x1024 (ablation) | PJD · Lumina-mGPT resolution sweep | step compression ratio: up to 6.9× | Improved |
Closest papers by meaning
- Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding (ICLR 2025 · similarity 0.93)
- ZipAR: Parallel Autoregressive Image Generation through Spatial Locality (ICML 2025 · similarity 0.91)
- Speculative Jacobi-Denoising Decoding for Accelerating Autoregressive Text-to-image Generation (NeurIPS 2025 · similarity 0.91)
- Hawk: Leveraging Spatial Context for Faster Autoregressive Text-to-Image Generation (NeurIPS 2025 · similarity 0.90)
- Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy (NeurIPS 2025 · similarity 0.90)
- Collaborative Decoding Makes Visual Auto-Regressive Modeling Efficient (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
