Multimodal Question Answering on MMBench English
85.7MMBenBaseline
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BaselineBackbone=InternVL3-8B, Retained Tokens=All 1280 Tokens, Reduction Percentage=100%2026.06 | 85.7 | 74.9 | 100 | |
| BaselineBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=All 1296 Tokens, Reduction Percentage=100%2026.06 | 84.8 | 73.1 | 100 | |
| TOPSBackbone=InternVL3-8B, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.0%2026.06 | 84.8 | 72.4 | 96.7 | |
| FastVBackbone=InternVL3-8B, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.0%2026.06 | 83.6 | 71.2 | 95.1 | |
| VisionZipBackbone=InternVL3-8B, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.0%2026.06 | 82 | 67.8 | 90.5 | |
| TOPSBackbone=InternVL3-8B, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.0%2026.06 | 82 | 68.8 | 91.9 | |
| DivPruneBackbone=InternVL3-8B, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.0%2026.06 | 81.9 | 69.3 | 92.5 | |
| DivPruneBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.2%2026.06 | 81.8 | 68.8 | 94.1 | |
| CDPrunerBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.2%2026.06 | 81.5 | 68.9 | 94.3 | |
| TOPSBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.2%2026.06 | 81.4 | 70.4 | 96.3 | |
| FastVBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.2%2026.06 | 80.5 | 67.8 | 92.7 | |
| TOPSBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.1%2026.06 | 80.2 | 67.9 | 92.9 | |
| DivPruneBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.1%2026.06 | 79.2 | 65.5 | 89.6 | |
| CDPrunerBackbone=InternVL3-8B, Retained Tokens=256 Tokens, Reduction Percentage=↓ 80.0%2026.06 | 79.2 | 67.4 | 90 | |
| CDPrunerBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.1%2026.06 | 78.6 | 65.5 | 89.6 | |
| DivPruneBackbone=InternVL3-8B, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.0%2026.06 | 78.3 | 64.6 | 86.2 | |
| CDPrunerBackbone=InternVL3-8B, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.0%2026.06 | 75.3 | 62.8 | 83.8 | |
| VisionZipBackbone=InternVL3-8B, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.0%2026.06 | 74.7 | 60.3 | 80.5 | |
| FastVBackbone=InternVL3-8B, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.0%2026.06 | 73.7 | 60.4 | 80.6 | |
| CoVFTBackbone=LLaVA-1.5-13B2026.03 | 70.45 | — | — | |
| BitFitBackbone=LLaVA-1.5-13B2026.03 | 70.02 | — | — | |
| FreezeBackbone=LLaVA-1.5-13B2026.03 | 69.93 | — | — | |
| Full fine-tuningBackbone=LLaVA-1.5-13B2026.03 | 68.56 | — | — | |
| CoVFTBackbone=LLaVA-1.5-7B2026.03 | 68.13 | — | — | |
| FreezeBackbone=LLaVA-1.5-7B2026.03 | 67.87 | — | — | |
| LLaVA-1.5Backbone=LLaVA-1.5-13B2026.03 | 67.7 | — | — | |
| Full fine-tuningBackbone=LLaVA-1.5-7B2026.03 | 67.44 | — | — | |
| LoRABackbone=LLaVA-1.5-7B2026.03 | 67.44 | — | — | |
| BitFitBackbone=LLaVA-1.5-7B2026.03 | 67.27 | — | — | |
| VPTBackbone=LLaVA-1.5-7B2026.03 | 67.1 | — | — | |
| SVPTBackbone=LLaVA-1.5-7B2026.03 | 66.58 | — | — | |
| FastVBackbone=Qwen2.5-VL-7B-Instruct, Retained Tokens=128 Tokens, Reduction Percentage=↓ 90.1%2026.06 | 66.4 | 57 | 78 | |
| LLaVA-1.5Backbone=LLaVA-1.5-7B2026.03 | 64.3 | — | — |