Research map / Image generation / Autoregressive and latent generation
Grouped Speculative Decoding for Autoregressive Image Generation
Junhyuk So, Juncheol Shin, Hyunho Kook, Eunhyeok Park
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 GSD with group size G=3 instead of SJD | Lumina-mGPT SJD · Parti-Prompt (Lumina-mGPT 7B) | NFE: 1035.3 -> 898.97 | Improved |
| use GSD with G=3 instead of SJD | Lumina-mGPT SJD · MS-COCO 2017 val (Lumina-mGPT 7B) | FID: 30.78 -> 31.50 | Hurt |
| increase GSD group size to G=50 | GSD · Parti-Prompt (Lumina-mGPT 7B) | latency speedup: 2.38x -> 4.65x | Improved |
| increase GSD group size from G=50 to G=100 | GSD · Parti-Prompt (Lumina-mGPT 7B) | CLIP score: 32.075 -> 31.975 | Hurt |
| increase GSD group size from G=3 to G=25 | GSD · MS-COCO 2017 val (Lumina-mGPT 7B) | FID: 31.50 -> 33.55 | Hurt |
| use GSD (G=50) instead of Addition (eps=3e-1) naive lossy SD | SJD with lossy acceptance · Parti-Prompt (Lumina-mGPT 7B) | CLIP score: 31.776 -> 32.075 | Improved |
| use GSD (G=25) instead of Amplify (k=4) naive lossy SD | SJD with lossy acceptance · MS-COCO 2017 val (Lumina-mGPT 7B) | FID: 40.05 -> 33.55 | Improved |
| add context-aware clustering by expert p(x) instead of static embedding distance (ablation) | GSD · Parti-Prompt clustering ablation (Lumina-mGPT 7B) | NFE: 913.80 -> 636.75 | Improved |
| cluster by expert p(x) rather than draft q(x) (ablation) | GSD · Parti-Prompt clustering ablation (Lumina-mGPT 7B) | CLIP score: 31.791 -> 32.075 | Improved |
| use dynamic clustering (ours) rather than plain expert p(x) clustering (ablation) | GSD · Parti-Prompt clustering ablation (Lumina-mGPT 7B) | CLIP score: 32.056 -> 32.075 | No clear effect |
| use Jacobi Decoding instead of vanilla AR | Lumina-mGPT · Parti-Prompt (Lumina-mGPT 7B) | NFE: 2392 -> 2300.0 | No clear effect |
| use SJD instead of vanilla AR | Lumina-mGPT · Parti-Prompt (Lumina-mGPT 7B) | latency speedup: 1.00x -> 2.15x | Improved |
Closest papers by meaning
- Hawk: Leveraging Spatial Context for Faster Autoregressive Text-to-Image Generation (NeurIPS 2025 · similarity 0.92)
- LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding (ICLR 2025 · similarity 0.92)
- Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy (NeurIPS 2025 · similarity 0.90)
- Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding (ICLR 2025 · similarity 0.90)
- Speculative Jacobi-Denoising Decoding for Accelerating Autoregressive Text-to-image Generation (NeurIPS 2025 · similarity 0.90)
- Distilled Decoding 1: One-step Sampling of Image Auto-regressive Models with Flow Matching (ICLR 2025 · similarity 0.90)
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.
