Text-to-Image Generation on PartiPrompt
5.388Latency (s)SCD (πGS)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SCD (πGS)Backbone=Janus-Pro, L=322025.10 | 5.388 | — | — | — | 29.13 | 154.42 | |
| SCD (πMC)Backbone=Janus-Pro, L=322025.10 | 5.471 | — | — | — | 28.95 | 149.53 | |
| SCD (πMC)Backbone=Janus-Pro, L=162025.10 | 6.22 | — | — | — | 28.99 | 186.48 | |
| SJD++Backbone=Janus-Pro (1B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 6.27 | 2.29 | 2.48 | 0.2714 | 32.63 | — | |
| SCD (πGS)Backbone=Janus-Pro, L=162025.10 | 6.336 | — | — | — | 29.03 | 189.99 | |
| SJDBackbone=Janus-Pro (1B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 8.05 | 1.79 | 1.87 | 0.2731 | 32.75 | — | |
| SJDBackbone=Janus-Pro, L=162025.10 | 9.96 | — | — | — | 29.02 | 312 | |
| SJDBackbone=Janus-Pro, L=322025.10 | 10.25 | — | — | — | 28.97 | 308.17 | |
| Janus-ProModel Parameters=1B, Decoding Strategy=Autoregressive2025.12 | 14.39 | 1 | 1 | 0.2752 | 32.88 | — | |
| Vanilla ARBackbone=Janus-Pro2025.10 | 16.84 | — | — | — | 28.98 | 576 | |
| SJD++Backbone=LlamaGen (0.8B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 19.72 | 2.32 | 3.21 | 0.2638 | 31.15 | — | |
| SCD (πGS)Configuration=D, L=642025.10 | 23.42 | 4.79 | 4.36 | — | 32.089 | 548.75 | |
| SCD (πMC)Configuration=D, L=642025.10 | 23.58 | 4.76 | 4.36 | — | 32.113 | 548.26 | |
| GSDConfiguration=C, G=502025.10 | 25.24 | 4.45 | 3.76 | — | 32.075 | 636.75 | |
| LANTERN-VPConfiguration=LANTERN-VP2026.03 | 25.97 | 3.06 | 4.12 | — | 32.11 | 580.47 | |
| SCD (πMC)Configuration=D, L=322025.10 | 26.01 | 4.32 | 3.69 | — | 32.09 | 647.08 | |
| SCD (πGS)Configuration=D, L=322025.10 | 26.33 | 4.26 | 3.68 | — | 32.102 | 649.54 | |
| SJD-VPConfiguration=SJD-VP2026.03 | 27 | 2.94 | 3.32 | — | 32.15 | 720.65 | |
| SJD++Backbone=Lumina-mGPT (7B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 27.91 | 3.12 | 6.4 | 0.2873 | 32.07 | — | |
| SJD++Backbone=SimpleAR (0.5B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 29.68 | 2.82 | 3.34 | 0.1871 | 29 | — | |
| GSD-VPConfiguration=GSD-VP2026.03 | 31.1 | 2.55 | 3.56 | — | 32.13 | 671.6 | |
| LANTERNConfiguration=LANTERN(Jang et al., 2024b)2026.03 | 31.4 | 2.53 | 3.76 | — | 32.1 | 636.66 | |
| SJDBackbone=LlamaGen (0.8B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 32.45 | 1.41 | 2.09 | 0.266 | 31.28 | — | |
| SCD (πMC)Configuration=D, L=162025.10 | 32.86 | 3.42 | 2.92 | — | 32.12 | 817.29 | |
| SCD (πGS)Configuration=D, L=162025.10 | 32.92 | 3.41 | 2.91 | — | 32.07 | 820.12 | |
| GSDConfiguration=GSD (So et al., 2025b)2026.03 | 33.36 | 2.38 | 2.66 | — | 32.11 | 898.97 | |
| SJDConfiguration=SJD (Teng et al., 2024)2026.03 | 36.07 | 2.2 | 2.31 | — | 32.09 | 1,035.3 | |
| SJDConfiguration=B, L=162025.10 | 42.07 | 2.67 | 2.31 | — | 32.11 | 1,035.3 | |
| SJDBackbone=Lumina-mGPT (7B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 42.82 | 2.04 | 2.25 | 0.2895 | 32.1 | — | |
| SJDConfiguration=B, L=642025.10 | 42.99 | 2.61 | 2.31 | — | 32.095 | 1,036.2 | |
| SJDBackbone=SimpleAR (0.5B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 43.01 | 1.95 | 2.32 | 0.1894 | 29.1 | — | |
| SJDConfiguration=B, L=322025.10 | 43.27 | 2.6 | 2.3 | — | 32.087 | 1,038.2 | |
| LlamaGenModel Parameters=0.8B, Decoding Strategy=Autoregressive2025.12 | 45.67 | 1 | 1 | 0.2681 | 31.42 | — | |
| Lumina-mGPTConfiguration=Lumina-mGPT (Liu et al., 2024b)2026.03 | 79.37 | 1 | 1 | — | 32.09 | 2,392 | |
| Jacobi DecodingConfiguration=Jacobi Decoding (Song et al., 2021)2026.03 | 82.21 | 0.97 | 1.04 | — | 32.09 | 2,300 | |
| SimpleARModel Parameters=0.5B, Decoding Strategy=Autoregressive2025.12 | 83.62 | 1 | 1 | 0.1912 | 29.17 | — | |
| Lumina-mGPTModel Parameters=7B, Decoding Strategy=Autoregressive2025.12 | 87.25 | 1 | 1 | 0.2913 | 32.28 | — | |
| Vanilla ARConfiguration=A2025.10 | 112.29 | 1 | 1 | — | 32.091 | 2,392 | |
| SJD++Backbone=Emu3-Gen (8B), Decoding Strategy=Speculative Jacobi Decoding with token reuse2025.12 | 124.58 | 3.01 | 7.51 | 0.2123 | 29.83 | — | |
| SJDBackbone=Emu3-Gen (8B), Decoding Strategy=Speculative Jacobi Decoding2025.12 | 207.58 | 1.82 | 2.03 | 0.2142 | 30.3 | — | |
| Emu3-GenModel Parameters=8B, Decoding Strategy=Autoregressive2025.12 | 375.21 | 1 | 1 | 0.2164 | 30.42 | — |