Spatial Reasoning on Model Average
93.6Binary Robust AccuracyLoRA-DPO
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen3-VL-8B2026.06 | 93.6 | 87.9 | 81 | 3.7 | 7.5 | 11.9 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen3-VL-2B2026.06 | 93.3 | 86.7 | 79.9 | 9.4 | 14.1 | 16.1 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen2-VL-7B2026.06 | 91.8 | 84.6 | 79.5 | 0.4 | 1.9 | 3.1 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-v1.6-Vicuna-7B2026.06 | 88.5 | 78.7 | 62.8 | 17.4 | 32.6 | 33.4 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen2-VL-2B2026.06 | 87.5 | 76.5 | 60 | 14.4 | 25.5 | 28.2 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-v1.6-Mistral-7B2026.06 | 87.1 | 81 | 75.3 | 35.3 | 53.9 | 68.1 | |
| LoRA-DPOModel Family=InternVL family, Backbone=InternVL2.5-1B2026.06 | 81.6 | 70.1 | 62.2 | 10.2 | 26.2 | 28.5 | |
| LoRA-DPOModel Family=InternVL family, Backbone=InternVL2-1B2026.06 | 81.1 | 62.2 | 40.4 | 18.2 | 28.4 | 20.2 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-1.5-7B2026.06 | 75 | 64.1 | 52.9 | 41 | 53.5 | 49.4 |