Multi-modal Understanding on LLaVA Multi-modal Evaluation Suite v1.6 (test)
100Average ScoreVanilla 13B
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Vanilla 13BCount=28802024.12 | 100 | 65.4 | 70 | 1,901 | 86.2 | 73.5 | 81.8 | 64.3 | 36.2 | 71.9 | 129.4 | 2,506 | |
| VisionZip🔥Base Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 640, Mode=Fine-tuned2025.08 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 640, Mode=Fine-tuned2025.08 | 98.8 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=1922025.08 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=192, Mode=Fine-tuned2025.08 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=1922025.08 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 6402025.08 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=192, Mode=Fine-tuned2025.08 | 98.4 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=1922025.08 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 6402025.08 | 98.2 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=1922025.08 | 98 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=1922025.08 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 67%, Remain Token=1922025.08 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=1282025.08 | 97.8 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 320, Mode=Fine-tuned2025.08 | 97.8 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=128, Mode=Fine-tuned2025.08 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 6402025.08 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 320, Mode=Fine-tuned2025.08 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip 13BCount=6402024.12 | 97.5 | 63 | 68.6 | 1,871 | 85.7 | 71.2 | 79.7 | 62.2 | 36.4 | 68.8 | 48.2 | 1,219 | |
| MMTokBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=1282025.08 | 97.5 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 6402025.08 | 97.5 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=128, Mode=Fine-tuned2025.08 | 97.4 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 3202025.08 | 97.3 | — | — | — | — | — | — | — | — | — | — | — | |
| Vanilla 7BCount=28802024.12 | 97.2 | 64.2 | 67.9 | 1,842 | 86.4 | 70.2 | 80.1 | 61.3 | 35.1 | 70.2 | 54.2 | 1,598 | |
| DivPruneBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 6402025.08 | 97.1 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 78%, Remain Token=Up. 6402025.08 | 97.1 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=1282025.08 | 97 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=1282025.08 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=1282025.08 | 96.9 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 78%, Remain Token=1282025.08 | 96.8 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=642025.08 | 96.6 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=642025.08 | 96.4 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 3202025.08 | 96.4 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=642025.08 | 95.3 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 3202025.08 | 95.1 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 1602025.08 | 95.1 | — | — | — | — | — | — | — | — | — | — | — | |
| MMTokBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 1602025.08 | 95.1 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=64, Mode=Fine-tuned2025.08 | 95 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 160, Mode=Fine-tuned2025.08 | 95 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=642025.08 | 94.8 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=64, Mode=Fine-tuned2025.08 | 94.8 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip 13BCount=3202024.12 | 94.7 | 60.7 | 67.2 | 1,805 | 82 | 70.3 | 76.8 | 60.9 | 35.6 | 65.2 | 30.3 | 995 | |
| VisionZipBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 3202025.08 | 94.7 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip🔥Base Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 160, Mode=Fine-tuned2025.08 | 94.6 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 3202025.08 | 94.5 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 89%, Remain Token=Up. 3202025.08 | 94.5 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5-13B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=642025.08 | 93.7 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-1.5-7B (2023), Original Tokens=576 tokens, Compress Ratio=↓ 89%, Remain Token=642025.08 | 93.2 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 1602025.08 | 92.4 | — | — | — | — | — | — | — | — | — | — | — | |
| DivPruneBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 1602025.08 | 92 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZipBase Model=LLaVA-NeXT-13B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 1602025.08 | 91.4 | — | — | — | — | — | — | — | — | — | — | — | |
| VisionZip 13BCount=1602024.12 | 91.3 | 57.8 | 64.9 | 1,739 | 76.6 | 69.3 | 72.4 | 58.4 | 37 | 61.1 | 23.9 | 888 | |
| VisionZipBase Model=LLaVA-NeXT-7B (2024a), Original Tokens=Upper(Up.) 2880 tokens, Compress Ratio=↓ 94%, Remain Token=Up. 1602025.08 | 90.4 | — | — | — | — | — | — | — | — | — | — | — |