Visual Question Answering on MMBench (dev)
60.1Overall ScoreLLaVA-RLHF_13B×336
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| LLaVA-RLHF_13B×336Data=280k2023.09 | 60.1 | 29.2 | 67.2 | 56.5 | 60.9 | 53.8 | 71.5 | — | — | — | |
| Kosmos-2Data=14M2023.09 | 59.2 | 46.7 | 55.7 | 43.5 | 64.3 | 49 | 72.5 | — | — | — | |
| Shikra_7BData=5.5M2023.09 | 58.8 | 25.8 | 56.7 | 58.3 | 57.2 | 57.9 | 75.8 | — | — | — | |
| LLaVA-SFT+_13B×336Data=220k2023.09 | 57.5 | 25.8 | 65.7 | 54.8 | 57.9 | 51 | 68.5 | — | — | — | |
| IDEFICS_80BData=1M2023.09 | 54.6 | 29 | 67.8 | 46.5 | 56 | 48 | 61.9 | — | — | — | |
| LLaVA-SFT+_7BData=220k2023.09 | 52.1 | 28.3 | 63.2 | 37.4 | 53.2 | 35.9 | 66.8 | — | — | — | |
| Otter-I_9BData=2.8M2023.09 | 51.4 | 32.5 | 56.7 | 53.9 | 46.8 | 38.6 | 65.4 | — | — | — | |
| LLaVA-RLHF_7BData=280k2023.09 | 51.4 | 24.2 | 63.2 | 39.1 | 50.2 | 40 | 66.1 | — | — | — | |
| IDEFICS_9BData=1M2023.09 | 48.2 | 20.8 | 54.2 | 33 | 47.8 | 36.6 | 67.1 | — | — | — | |
| LLaVA_13B×336Data=158k2023.09 | 47.5 | 23.3 | 59.7 | 31.3 | 41.4 | 38.6 | 65.8 | — | — | — | |
| InstructBLIP_13BData=1.2M2023.09 | 44 | 19.1 | 54.2 | 34.8 | 47.8 | 24.8 | 56.4 | — | — | — | |
| LLaMA-Adapter_7BData=52k2023.09 | 41.2 | 11.7 | 35.3 | 29.6 | 47.5 | 38.6 | 56.4 | — | — | — | |
| LLaVA_7BData=158k2023.09 | 38.7 | 16.7 | 48.3 | 30.4 | 45.5 | 32.4 | 40.6 | — | — | — | |
| InstructBLIP_7BData=1.2M2023.09 | 36 | 14.2 | 46.3 | 22.6 | 37 | 21.4 | 49 | — | — | — | |
| MiniGPT-4_7BData=5k2023.09 | 24.3 | 7.5 | 31.3 | 4.3 | 30.3 | 9 | 35.6 | — | — | — | |
| OpenFlamingo_9BData=not specified2023.09 | 6.6 | 4.2 | 15.4 | 0.9 | 8.1 | 1.4 | 5 | — | — | — | |
| CLIP-ScoreTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 55.2 | 52 | |
| COIDOTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 63.8 | 56.7 | |
| COINCIDETraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 63.1 | 54.5 | |
| D2-PruningTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 65.7 | 57.6 | |
| DeepSeekVL-2Model Size=2B2025.01 | — | — | — | — | — | — | — | 68.3 | — | — | |
| DeepSeekVL-2Model Size=8B2025.01 | — | — | — | — | — | — | — | 78.1 | — | — | |
| EL2NTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 53.2 | 47.4 | |
| FullTraining Data Fraction=100%2026.05 | — | — | — | — | — | — | — | — | 66.1 | 58.9 | |
| ICONSTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 63.1 | 55.8 | |
| InternVL2.5-MPOModel Size=2B2025.01 | — | — | — | — | — | — | — | 72.5 | — | — | |
| InternVL2.5-MPOModel Size=8B2025.01 | — | — | — | — | — | — | — | 83.3 | — | — | |
| MAGICTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 65.1 | 56.8 | |
| PerplexityTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 52 | 45.8 | |
| Qwen2-VLModel Size=2B2025.01 | — | — | — | — | — | — | — | 68.8 | — | — | |
| Qwen2-VLModel Size=8B2025.01 | — | — | — | — | — | — | — | 79.5 | — | — | |
| RandomTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 62.2 | 54.8 | |
| RDSTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 61.2 | 52.7 | |
| SAIL-VLModel Size=2B2025.01 | — | — | — | — | — | — | — | 72.4 | — | — | |
| SAIL-VLModel Size=8B2025.01 | — | — | — | — | — | — | — | 79.5 | — | — | |
| Self-FilterTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 48.8 | 45.3 | |
| Self-SepTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 61.4 | 53.8 | |
| SemDeDupTraining Data Fraction=20%2026.05 | — | — | — | — | — | — | — | — | 52.2 | 48.5 |