User Preference & Fluency on MMVet
41.5MMVet User Preference ScoreScaled base VLM 7B + Joint FFT
Evaluation results
| Method | Links | |
|---|---|---|
| Scaled base VLM 7B + Joint FFTTrain Data=3.6M + 1.6M2026.06 | 41.5 | |
| Scaled base VLM 7B + MERIT (Proposed, 2D)Train Data=3.6M + 1.6M2026.06 | 39.1 | |
| LLaVA-1.5-13BTrain Data=0.7M2026.06 | 36.1 | |
| Base VLM 7B + MERIT (Proposed, 2D)Train Data=0.7M + 1.6M2026.06 | 35.1 | |
| Scaled base VLM 7BTrain Data=3.6M2026.06 | 34 | |
| LLaVA-1.5-13B + Joint FFTTrain Data=0.7M + 0.2M2026.06 | 33.4 | |
| Base VLM 7B + Joint FFTTrain Data=0.7M + 1.6M2026.06 | 32.5 | |
| LLaVA-1.5-7BTrain Data=0.7M2026.06 | 31.1 | |
| Base VLM 7BTrain Data=0.7M2026.06 | 28.9 | |
| LLaVA-7BTrain Data=0.6M2026.06 | 25.5 |