List your company

Geometric Reasoning on GeoQA

72.52AccuracyGeoSketch-Qwen2.5VL-7B
21.268834.574447.8861.1856Sep 26, 2025Nov 2, 2025Dec 10, 2025Jan 17, 2026Feb 23, 2026Apr 2, 2026May 10, 2026

Evaluation results

MethodLinks
GeoSketch-Qwen2.5VL-7BTraining Stage=RL2025.0972.52
Theorem-SFTBackbone=Qwen2.5-VL-7B, Supervision Layers=MLP2026.0569.73
Theorem-SFTBackbone=Qwen2.5-VL-7B, Supervision Layers=All2026.0566.89
GeoSketch-Qwen2.5VL-7BTraining Stage=SFT2025.0965.09
BaseBackbone=Qwen2.5-VL-7B2026.0549.46
Qwen2.5VL-7BMode=Base2025.0947.28
Theorem-SFTBackbone=Qwen2.5-VL-3B, Supervision Layers=MLP2026.0541.89
Vanilla SFTBackbone=Qwen2.5-VL-7B2026.0538.78
Theorem-SFTBackbone=Gemma-3-4B-IT, Supervision Layers=MLP2026.0536.72
BaseBackbone=Qwen2.5-VL-3B2026.0535.95
Theorem-SFTBackbone=Gemma-3-4B-IT, Supervision Layers=All2026.0535.72
Theorem-SFTBackbone=Qwen2.5-VL-3B, Supervision Layers=All2026.0535.54
Vanilla SFTBackbone=Qwen2.5-VL-3B2026.0534.59
Vanilla SFTBackbone=Gemma-3-4B-IT2026.0531.89
BaseBackbone=Gemma-3-4B-IT2026.0523.24