Paired-prompt evaluation on MME
75.7Simple AccuracyLLaVA-NeXT-Vicuna-7B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LLaVA-NeXT-Vicuna-7BIntervention=No-intervention baseline2026.01 | 75.7 | 15.38 | 56.44 | |
| LLaVA-NeXT-Vicuna-7BIntervention=Q4 system redistribution (prop)2026.01 | 40.61 | 0.77 | 20.56 |