Visual Perception on V* v1.0 (test)
84.35ScoreFull Cache
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full CacheBackbone=InternVL3.5-8B, KV Budget=Full2026.04 | 84.35 | 67.46 | — | |
| Full CacheBackbone=Qwen2.5-VL-32B, KV Budget=Full2026.04 | 83.77 | 70.99 | — | |
| SnapKVBackbone=InternVL3.5-8B, KV Budget=20482026.04 | 83.76 | 66.5 | 0.96 | |
| Full CacheBackbone=InternVL3.5-14B, KV Budget=Full2026.04 | 83.76 | 68.95 | — | |
| SnapKVBackbone=InternVL3.5-8B, KV Budget=40962026.04 | 82.72 | 66.5 | 0.96 | |
| SnapKVBackbone=InternVL3.5-14B, KV Budget=10242026.04 | 82.72 | 63.26 | 7.73 | |
| SnapKVBackbone=Qwen2.5-VL-32B, KV Budget=10242026.04 | 82.72 | 63.26 | 7.73 | |
| SnapKVBackbone=InternVL3.5-8B, KV Budget=10242026.04 | 82.61 | 66.68 | 0.78 | |
| Full CacheBackbone=Qwen2.5-VL-7B, KV Budget=Full2026.04 | 79.58 | 65.52 | — | |
| KeyDiffBackbone=Qwen2.5-VL-7B, KV Budget=40962026.04 | 79.58 | 64.47 | 1.05 | |
| SnapKVBackbone=Qwen2.5-VL-7B, KV Budget=40962026.04 | 78.53 | 64.28 | 1.24 | |
| SnapKVBackbone=Qwen2.5-VL-7B, KV Budget=20482026.04 | 78.53 | 61.31 | 4.21 | |
| SnapKVBackbone=Qwen2.5-VL-7B, KV Budget=10242026.04 | 76.96 | 54.46 | 11.06 | |
| KeyDiffBackbone=InternVL3.5-8B, KV Budget=20482026.04 | 75.39 | 65.23 | 2.23 | |
| KeyDiffBackbone=InternVL3.5-8B, KV Budget=40962026.04 | 74.87 | 64.03 | 3.43 | |
| KeyDiffBackbone=InternVL3.5-8B, KV Budget=10242026.04 | 74.35 | 62.76 | 4.7 | |
| KeyDiffBackbone=Qwen2.5-VL-32B, KV Budget=10242026.04 | 72.25 | 59.96 | 11.03 | |
| KeyDiffBackbone=Qwen2.5-VL-7B, KV Budget=20482026.04 | 69.63 | 60.62 | 4.9 | |
| KeyDiffBackbone=Qwen2.5-VL-7B, KV Budget=10242026.04 | 67.02 | 55.82 | 9.7 | |
| KeyDiffBackbone=InternVL3.5-14B, KV Budget=10242026.04 | 64.92 | 60.01 | 8.84 |