Visual Question Answering on ScienceQA (val)
82.9AccuracySNs (mean)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SNs (mean)Backbone=Qwen3-VL-4b-Instruct2026.03 | 82.9 | 82 | 71.7 | 78.3 | 74.8 | |
| SNs (maj. voting)Backbone=Qwen3-VL-4b-Instruct2026.03 | 82.4 | 82 | 72.9 | 73.4 | 73.2 | |
| Qwen3-VL-4b-Instruct2026.03 | 81.1 | — | 68.3 | 78.7 | 73.1 | |
| Uni-DPOModel=Qwen2-VL-2B2025.06 | 68.1 | — | — | — | — | |
| SimPOModel=Qwen2-VL-2B2025.06 | 66.81 | — | — | — | — | |
| BaselineModel=Qwen2-VL-2B2025.06 | 65.57 | — | — | — | — | |
| DPOModel=Qwen2-VL-2B2025.06 | 63 | — | — | — | — | |
| SNs (mean)Backbone=LLaVA-v1.5-7b2026.03 | 57.4 | 63 | 42.4 | 84.5 | 56.5 | |
| LLaVA-v1.5-7b2026.03 | 53.5 | — | 40.1 | 85.9 | 54.7 | |
| SNs (maj. voting)Backbone=LLaVA-v1.5-7b2026.03 | 48.4 | 63 | 38.3 | 94.9 | 54.6 |