Spatial Reasoning on VSR (Visual Spatial Reasoning)
75.8Binary Robust AccLoRA-DPO
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen3-VL-8B2026.06 | 75.8 | 57.7 | 42.6 | 2.4 | 3.2 | 4.5 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen3-VL-2B2026.06 | 75.3 | 58.4 | 43.9 | 5.9 | 7.5 | 9.3 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen2-VL-7B2026.06 | 69.7 | 49 | 35.8 | -1.1 | 0 | 3 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen2-VL-2B2026.06 | 67.9 | 48.9 | 33.6 | 6.2 | 9.2 | 10.1 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-v1.6-Vicuna-7B2026.06 | 65.9 | 42.1 | 23.9 | 10.1 | 10.9 | 6 | |
| LoRA-DPOModel Family=InternVL family, Backbone=InternVL2-1B2026.06 | 59.2 | 34.1 | 15 | 3.1 | 8.8 | 6.1 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-v1.6-Mistral-7B2026.06 | 58.6 | 43.1 | 29.5 | 14.9 | 22.3 | 20.1 | |
| LoRA-DPOModel Family=InternVL family, Backbone=InternVL2.5-1B2026.06 | 50.8 | 31.9 | 22 | 3 | 10 | 8.3 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-1.5-7B2026.06 | 49 | 28.4 | 16.6 | 12.5 | 11.6 | 7.7 |