Spatial Reasoning on SpatialMQA Direct
91.8Binary Robust AccuracyLoRA-DPO
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen3-VL-8B2026.06 | 91.8 | 83.9 | 76 | 1.3 | 2.2 | 4.6 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen3-VL-2B2026.06 | 87.2 | 76.6 | 66 | 1.2 | 2 | 4.1 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen2-VL-7B2026.06 | 81.9 | 67.3 | 55.7 | -0.7 | -0.3 | 0.5 | |
| LoRA-DPOModel Family=Qwen family, Backbone=Qwen2-VL-2B2026.06 | 72.4 | 57.5 | 41.9 | -1.3 | 2.9 | 2.1 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-v1.6-Vicuna-7B2026.06 | 70.7 | 51.6 | 32.8 | 13.8 | 25.2 | 23.1 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-v1.6-Mistral-7B2026.06 | 70.1 | 58.7 | 48.2 | 8.7 | 21.4 | 32.6 | |
| LoRA-DPOModel Family=InternVL family, Backbone=InternVL2-1B2026.06 | 59.4 | 38.8 | 24.9 | -3 | 2.6 | 4.7 | |
| LoRA-DPOModel Family=InternVL family, Backbone=InternVL2.5-1B2026.06 | 54.9 | 35.4 | 28.6 | 0.1 | 8.8 | 3.6 | |
| LoRA-DPOModel Family=LLaVA family, Backbone=LLaVA-1.5-7B2026.06 | 49.6 | 33.6 | 18.8 | 33.4 | 28.2 | 16.2 |