Multi-modal preference alignment on MM-AlignBench
62.3Winning RateLLaVANext
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLaVANextData=OmniAlign-Vmix, LLM=Qwen2.5-32B2025.02 | 62.3 | 19.4 | |
| LLaVANextData=OmniAlign-Vmix, LLM=InternLM2.5-7B2025.02 | 57.1 | 11.1 | |
| LLaVAData=OmniAlign-Vmix, LLM=InternLM2.5-7B2025.02 | 50 | 3.8 | |
| LLaVANextData=LLaVANext-778k, LLM=Qwen2.5-32B2025.02 | 26.6 | -29 | |
| LLaVANextData=LLaVANext-778k, LLM=InternLM2.5-7B2025.02 | 20.6 | -42.7 | |
| LLaVAData=LLaVANext-778k, LLM=InternLM2.5-7B2025.02 | 3.6 | -82.1 |