Math Reasoning on We-Math
85.6Avg Pass@8ADHint
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ADHintBase Model=Qwen3-VL-8B-Instruct2025.12 | 85.6 | 76.4 | |
| GRPOBase Model=Qwen3-VL-8B-Instruct2025.12 | 82.6 | 73.1 | |
| SFT -> RLBase MLLM=Qwen2.5-VL-7B2025.12 | 72.5 | 54.1 | |
| HintGRPOBase Model=Qwen3-VL-8B-Instruct2025.12 | 71.7 | 68.6 | |
| ADHintBase MLLM=Qwen2.5-VL-7B2025.12 | 70.8 | 69.9 | |
| SFTBase MLLM=Qwen2.5-VL-7B2025.12 | 70.1 | 44.1 | |
| GRPOBase MLLM=Qwen2.5-VL-7B2025.12 | 68.8 | 69.8 | |
| SFT -> RLBase MLLM=Qwen2.5-VL-3B2025.12 | 60.6 | 38.3 | |
| ADHintBase MLLM=Qwen2.5-VL-3B2025.12 | 59.1 | 60.4 | |
| GRPOBase MLLM=Qwen2.5-VL-3B2025.12 | 59 | 59 | |
| SFTBase MLLM=Qwen2.5-VL-3B2025.12 | 58.5 | 32.5 | |
| LUFFYBase MLLM=Qwen2.5-VL-7B2025.12 | 54.8 | 52.2 | |
| GHPOBase MLLM=Qwen2.5-VL-7B2025.12 | 52.8 | 63 | |
| VA-OPDTeacher=Qwen3-VL-8B, Student=Qwen3-VL-2B, Training Data=Geometry3K, Temperature=1.0, Checkpoint Selection=Best checkpoint2026.05 | 46.6 | — | |
| LUFFYBase MLLM=Qwen2.5-VL-3B2025.12 | 46.1 | 42.3 | |
| PAPOTeacher=Qwen3-VL-8B, Student=Qwen3-VL-2B, Training Data=Geometry3K, Temperature=1.0, Checkpoint Selection=Best checkpoint2026.05 | 44.8 | — | |
| HintGRPOBase MLLM=Qwen2.5-VL-7B2025.12 | 44.7 | 42.9 | |
| GRPOTeacher=Qwen3-VL-8B, Student=Qwen3-VL-2B, Training Data=Geometry3K, Temperature=1.0, Checkpoint Selection=Best checkpoint2026.05 | 44.1 | — | |
| OPDTeacher=Qwen3-VL-8B, Student=Qwen3-VL-2B, Training Data=Geometry3K, Temperature=1.0, Checkpoint Selection=Best checkpoint2026.05 | 43.3 | — | |
| StepHintBase MLLM=Qwen2.5-VL-7B2025.12 | 41.1 | 39.8 | |
| HintGRPOBase MLLM=Qwen2.5-VL-3B2025.12 | 40.5 | 42 | |
| StepHintBase MLLM=Qwen2.5-VL-3B2025.12 | 40.2 | 32 | |
| CoT-SFTTeacher=Qwen3-VL-8B, Student=Qwen3-VL-2B, Training Data=Geometry3K, Temperature=1.0, Checkpoint Selection=Best checkpoint2026.05 | 39.7 | — | |
| GHPOBase MLLM=Qwen2.5-VL-3B2025.12 | 39.1 | 41 | |
| Off-policy KDTeacher=Qwen3-VL-8B, Student=Qwen3-VL-2B, Training Data=Geometry3K, Temperature=1.0, Checkpoint Selection=Best checkpoint2026.05 | 38.9 | — | |
| BaseTeacher=Qwen3-VL-8B, Student=Qwen3-VL-2B, Training Data=Geometry3K, Temperature=1.0, Checkpoint Selection=Best checkpoint2026.05 | 36.8 | — |