Helpfulness Evaluation on LLaVA-Bench
93.1Conversation ScoreLLaVA-RLHF
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaVA-RLHFModel Size=13B2025.01 | 93.1 | 76.2 | 105.6 | 91.8 | |
| RLHF-VModel Size=13B2025.01 | 93.1 | 75.3 | 91.6 | 86.7 | |
| OPA-DPOModel Size=13B2025.01 | 87.1 | 78.3 | 90.7 | 85.5 | |
| POVIDModel Size=7B2025.01 | 84.9 | 77.3 | 90.3 | 84.3 | |
| LLaVA-Instruct-1.5-7BModel Size=7B2025.01 | 84.1 | 74.4 | 89.8 | 83 | |
| LLaVA-RLHFModel Size=7B2025.01 | 84.1 | 75.3 | 106.8 | 88.9 | |
| OPA-DPOModel Size=7B2025.01 | 82.1 | 79.5 | 87.9 | 83.2 | |
| HA-DPOModel Size=7B2025.01 | 80.7 | 74.5 | 88.4 | 81.4 | |
| LLaVA-Instruct-1.5-13BModel Size=13B2025.01 | 79.6 | 77.3 | 91.4 | 82.9 | |
| HSA-DPOModel Size=13B2025.01 | 76 | 71.8 | 88.2 | 80.5 | |
| RLAIF-VModel Size=7B2025.01 | 75.8 | 83.7 | 90.7 | 83.5 |