Image Generation on ImageNet 256x256 (CFG 1.0 & 1.5)
9.87FID (CFG=1.0)DSMoE-L-E48
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DSMoE-L-E48#Activated Params.=436M, #Total Params.=1.112B2025.12 | 9.87 | 111.54 | 2.72 | 263.13 | |
| DSMoE-L-E16#Activated Params.=465M, #Total Params.=1.304B2025.12 | 10.71 | 107.34 | 2.78 | 258.09 | |
| ProMoE-L#Activated Params.=458M, #Total Params.=1.063B2025.12 | 11.61 | 100.82 | 2.79 | 244.21 | |
| ProMoE-L-Flow#Experts=E14A1S1U1, # Activated Params.=458M, # Total Params.=1.063B, # GFLOPs=77.722025.10 | 11.61 | 100.82 | 2.79 | 244.21 | |
| DiffMoE-L-Flow#Experts=E16A1S0U0, # Activated Params.=458M, # Total Params.=1.846B, # GFLOPs=90.032025.10 | 13.55 | 92.33 | 3.3 | 222.4 | |
| DiffMoE-L-Flow#Experts=E8A1S0U0, # Activated Params.=458M, # Total Params.=1.095B, # GFLOPs=82.532025.10 | 14.46 | 87.55 | 3.51 | 212.78 | |
| Dense-DiT-L#Activated Params.=458M, #Total Params.=458M2025.12 | 15.44 | 84.2 | 3.56 | 209.03 | |
| EC-DiT-L-Flow#Experts=E8A1S0U0, # Activated Params.=458M, # Total Params.=1.163B, # GFLOPs=77.502025.10 | 15.58 | 84.11 | 3.65 | 209.06 | |
| DiT-MoE-L-Flow#Experts=E8A1S0U0, # Activated Params.=458M, # Total Params.=1.163B, # GFLOPs=77.502025.10 | 16.57 | 80.25 | 4.1 | 199.05 | |
| DSMoE-B-E48#Activated Params.=118M, #Total Params.=263M2025.12 | 21.56 | 66.48 | 5.14 | 172.8 | |
| DSMoE-B-E16#Activated Params.=132M, #Total Params.=368M2025.12 | 22.46 | 64.57 | 5.38 | 170.22 | |
| ProMoE-B#Activated Params.=130M, #Total Params.=300M2025.12 | 24.44 | 60.38 | 6.39 | 154.21 | |
| Dense-DiT-B#Activated Params.=130M, #Total Params.=130M2025.12 | 30.61 | 49.89 | 9.02 | 131.13 |