Multimodal Understanding on MMBench (Score, Latency, Throughput, Memory)
6,023LatencyTopV
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TopVModel=LLaVA-v1.5-13B, FLOPs Ratio=50%2025.03 | 6,023 | 65.16 | 29.11 | 6.56 | — | |
| TopVModel=LLaVA-v1.5-13B, FLOPs Ratio=35%2025.03 | 6,358 | 65.63 | 29.41 | 6.32 | — | |
| TopVModel=LLaVA-v1.5-7B, FLOPs Ratio=51%2025.03 | 6,702 | 59.65 | 14.77 | 6.46 | — | |
| TopVModel=LLaVA-v1.5-7B, FLOPs Ratio=35%2025.03 | 7,236 | 60.42 | 14.9 | 6.4 | — | |
| BaselineModel=LLaVA-v1.5-13B, FLOPs Ratio=02025.03 | 7,438 | 65.46 | 30.06 | 5.43 | — | |
| TopVModel=InternVL2-2B, FLOPs Ratio=48%2025.03 | 7,517 | 69.7 | 5.98 | 14.57 | — | |
| BaselineModel=LLaVA-v1.5-7B, FLOPs Ratio=02025.03 | 7,750 | 59.97 | 15.05 | 5.29 | — | |
| FastVModel=LLaVA-v1.5-13B, FLOPs Ratio=48%2025.03 | 8,648 | 64.95 | 40.12 | 4.63 | — | |
| BaselineModel=InternVL2-2B, FLOPs Ratio=02025.03 | 8,657 | 67.6 | 6.78 | 12.1 | — | |
| Multi-Agent Inference2026.04 | 8,676 | — | — | — | 4.69 | |
| FastVModel=LLaVA-v1.5-7B, FLOPs Ratio=47%2025.03 | 9,801 | 59.62 | 21.14 | 4.2 | — | |
| FastVModel=InternVL2-2B, FLOPs Ratio=47%2025.03 | 13,656 | 69.46 | 7.62 | 7.68 | — | |
| TopVModel=InternVL2-26B, FLOPs Ratio=47%2025.03 | 30,052 | 81.27 | 56.33 | 4.83 | — | |
| BaselineModel=InternVL2-26B, FLOPs Ratio=02025.03 | 34,227 | 80.89 | 57.02 | 4.24 | — | |
| Full Reasoning Transfer2026.04 | 40,718 | — | — | — | — | |
| FastVModel=InternVL2-26B, FLOPs Ratio=46%2025.03 | 49,038 | 80.86 | 58.87 | 2.96 | — |