Multimodal Understanding on Standard Multimodal Evaluation Suite 1.5 (test val)
63.2GQA ScoreLLaVA-1.5-13B
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LLaVA-1.5-13BBase Model=LLaVA-1.5-13B, Retained Tokens=576, Token Reduction Percentage=0%2025.03 | 63.2 | 67.7 | 1,818 | 61.3 | 72.8 | 80 | 100 | |
| LLaVA-1.5-13BBase Model=LLaVA-1.5-13B, Retained Tokens=576, Token Reduction Percentage=100%2025.03 | 63.2 | 67.7 | 1,818 | 61.3 | 72.8 | 80 | 100 | |
| TwigVLM++Base Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 62.6 | 68.2 | 1,829 | 60.4 | 73.2 | 79.4 | 99.8 | |
| TwigVLM++Base Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 62.6 | 68.2 | 1,829 | 60.4 | 73.2 | 79.4 | 99.8 | |
| TwigVLMBase Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 62.5 | 68.6 | 1,840 | 60.4 | 73.1 | 79.4 | 99.9 | |
| TwigVLMBase Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 62.5 | 68.6 | 1,840 | 60.4 | 73.1 | 79.4 | 99.9 | |
| TwigVLM++Base Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 62.3 | 67.7 | 1,849 | 60.3 | 72.9 | 79.1 | 99.6 | |
| TwigVLM++Base Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 62.3 | 67.7 | 1,849 | 60.3 | 72.9 | 79.1 | 99.6 | |
| VisionZip‡Base Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 61.6 | 67.1 | 1,790 | 59.9 | 72.7 | 78.6 | 98.5 | |
| VisionZip‡Base Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 61.6 | 67.1 | 1,790 | 59.9 | 72.7 | 78.6 | 98.5 | |
| TwigVLMBase Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 61.2 | 66.9 | 1,811 | 60.2 | 73.4 | 79.1 | 98.9 | |
| TwigVLMBase Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 61.2 | 66.9 | 1,811 | 60.2 | 73.4 | 79.1 | 98.9 | |
| TwigVLM++Base Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 61 | 65.7 | 1,793 | 58.5 | 72.7 | 78 | 97.5 | |
| TwigVLM++Base Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 61 | 65.7 | 1,793 | 58.5 | 72.7 | 78 | 97.5 | |
| FastVBase Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 60.3 | 67.4 | 1,807 | 60.4 | 74 | 77.7 | 98.6 | |
| FastVBase Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 60.3 | 67.4 | 1,807 | 60.4 | 74 | 77.7 | 98.6 | |
| VisionZip‡Base Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 60.1 | 67.6 | 1,736 | 59.2 | 73 | 77.6 | 97.4 | |
| VisionZip‡Base Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 60.1 | 67.6 | 1,736 | 59.2 | 73 | 77.6 | 97.4 | |
| TwigVLMBase Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 60 | 67.4 | 1,765 | 58.4 | 72.4 | 77 | 97.1 | |
| TwigVLMBase Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 60 | 67.4 | 1,765 | 58.4 | 72.4 | 77 | 97.1 | |
| VisionZipBase Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 59.1 | 66.9 | 1,754 | 59.5 | 73.5 | 78.1 | 97.4 | |
| VisionZipBase Model=LLaVA-1.5-13B, Retained Tokens=192, Token Reduction Percentage=66.7%2025.03 | 59.1 | 66.9 | 1,754 | 59.5 | 73.5 | 78.1 | 97.4 | |
| VisionZip‡Base Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 58.1 | 65.6 | 1,671 | 58.5 | 72.3 | 75.2 | 94.9 | |
| VisionZip‡Base Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 58.1 | 65.6 | 1,671 | 58.5 | 72.3 | 75.2 | 94.9 | |
| VisionZipBase Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 57.9 | 66.7 | 1,743 | 58.7 | 74 | 76.8 | 96.6 | |
| VisionZipBase Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 57.9 | 66.7 | 1,743 | 58.7 | 74 | 76.8 | 96.6 | |
| FastVBase Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 57.5 | 65.9 | 1,758 | 58 | 73.8 | 74.3 | 95.7 | |
| FastVBase Model=LLaVA-1.5-13B, Retained Tokens=128, Token Reduction Percentage=77.8%2025.03 | 57.5 | 65.9 | 1,758 | 58 | 73.8 | 74.3 | 95.7 | |
| VisionZipBase Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 56.2 | 64.9 | 1,676 | 57.4 | 74.4 | 73.7 | 94.2 | |
| VisionZipBase Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 56.2 | 64.9 | 1,676 | 57.4 | 74.4 | 73.7 | 94.2 | |
| FastVBase Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 50.1 | 55.9 | 1,408 | 52.2 | 73.2 | 61.1 | 83.6 | |
| FastVBase Model=LLaVA-1.5-13B, Retained Tokens=64, Token Reduction Percentage=88.9%2025.03 | 50.1 | 55.9 | 1,408 | 52.2 | 73.2 | 61.1 | 83.6 |