Video Generation on VBench Wan2.1 14B, 93x704x1280 1.0 (test)
67.49AQFull Attention
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Full AttentionModel=Wan2.1 14B, Resolution=93x704x1280, Kernel=FullAttn, Training Setting=Fine-tuned2026.01 | 67.49 | 97.04 | 61.11 | 68.53 | 95.23 | — | 7,903.8 | 2,222.2 | |
| Full AttentionModel=Wan2.1 14B, Resolution=93x704x1280, Kernel=FullAttn, Training Setting=Training-Free2026.01 | 67.42 | 96.02 | 66.67 | 66.73 | 93.52 | — | 7,903.8 | 2,222.2 | |
| VSAModel=Wan2.1 14B, Resolution=93x704x1280, Kernel=VSA, Training Setting=Fine-tuned2026.01 | 66.4 | 96.93 | 59.72 | 65.91 | 94.05 | 90 | 2,642.7 | 970.9 | |
| VMonarchModel=Wan2.1 14B, Resolution=93x704x1280, Kernel=VMonarch, Training Setting=Fine-tuned2026.01 | 65.91 | 97.32 | 58.33 | 66.61 | 95.68 | 92 | 2,670.5 | 969.2 |