Image-to-Video Generation on COCO 2017
0.79VBench-QFlashAttention (baseline)
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FlashAttention (baseline)Sparsification=no sparsification, Steps=42026.06 | 0.79 | 0.922 | 24.12 | — | — | — | — | 170.1 | 1 | |
| SageAttentionSteps=42026.06 | 0.789 | 0.922 | 24.05 | 24.817 | 0.852 | 0.093 | 0.9894 | 91.8 | 1.85 | |
| Sparge-Attntop-k=0.35, Steps=42026.06 | 0.788 | 0.922 | 24.1 | 20.879 | 0.741 | 0.169 | 0.9791 | 76.2 | 2.23 | |
| Sparge-Attntop-k=0.5, Steps=42026.06 | 0.788 | 0.921 | 24.11 | 21.481 | 0.772 | 0.153 | 0.9824 | 80.5 | 2.11 | |
| SVG2mode=fast, top-p=0.9, Steps=42026.06 | 0.788 | 0.921 | 24.1 | 22.213 | 0.791 | 0.135 | 0.984 | 135.6 | 1.25 | |
| Ours (sparsity reuse, turbo)Steps=4, top_p=0.9, min_top_k=0.02026.06 | 0.788 | 0.923 | 24.2 | 21.128 | 0.759 | 0.163 | 0.9802 | 75.6 | 2.25 | |
| SVG2mode=default, top-p=0.95, Steps=42026.06 | 0.787 | 0.917 | 24.14 | 23.124 | 0.817 | 0.118 | 0.9866 | 160 | 1.06 | |
| Ours (sparsity + feature reuse)Steps=4, Feature Reuse=true2026.06 | 0.787 | 0.924 | 24.5 | 13.211 | 0.5 | 0.461 | 0.9305 | 65.7 | 2.59 | |
| Ours (sparsity reuse, fp16)Steps=4, Precision=fp162026.06 | 0.787 | 0.918 | 24.11 | 23.273 | 0.818 | 0.119 | 0.986 | 129.8 | 1.31 | |
| Ours (sparsity reuse)Steps=4, top_p=0.95, min_top_k=0.12026.06 | 0.787 | 0.918 | 24.16 | 23.252 | 0.818 | 0.12 | 0.9859 | 78.9 | 2.16 | |
| Radial-AttnSteps=42026.06 | 0.786 | 0.92 | 24.14 | 20.043 | 0.723 | 0.194 | 0.9761 | 130 | 1.31 | |
| Sparge-Attntop-k=0.08, Steps=42026.06 | 0.779 | 0.894 | 24.11 | 16.157 | 0.591 | 0.324 | 0.9499 | 68.3 | 2.49 |