Video Generation on VBench HunyuanVideo-T2V-13B
62.1Sparsity (%)VecAttention
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VecAttentionResolution=720P, #Tokens (Context Length)=119K, Inference steps=50, Warm-up ratio=6%2026.03 | 62.1 | 22.8 | 77.9 | 0.33 | |
| XAttentionResolution=720P, #Tokens (Context Length)=119K, Inference steps=50, Warm-up ratio=6%2026.03 | 60.8 | 21.2 | 73.4 | 0.348 | |
| SVGResolution=720P, #Tokens (Context Length)=119K, Inference steps=50, Warm-up ratio=6%2026.03 | 60.1 | 21.8 | 76.9 | 0.326 |