Video Generation on Wan2.1-1.3B 4-step distilled
0.846VBenchDense Attention
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Dense AttentionInference steps=4-step2026.02 | 0.846 | — | — | — | |
| MONARCHRTSparsity level=90%, Training-free evaluation=true, Inference steps=4-step2026.02 | 0.834 | 12.657 | 0.364 | 0.585 | |
| SVG2Sparsity level=~85%, Training-free evaluation=true, Inference steps=4-step2026.02 | 0.823 | 11.154 | 0.321 | 0.631 | |
| SVG2Sparsity level=~90%, Training-free evaluation=true, Inference steps=4-step2026.02 | 0.808 | 10.737 | 0.307 | 0.662 | |
| SVGSparsity level=85%, Training-free evaluation=true, Inference steps=4-step2026.02 | 0.746 | 9.411 | 0.203 | 0.749 | |
| RadialAttentionSparsity level=85%, Training-free evaluation=true, Inference steps=4-step2026.02 | 0.727 | 11.427 | 0.29 | 0.711 |