Language Understanding on Average In-domain Agieval MMLU
65.3AccuracyRLVR
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RLVRBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 65.3 | 0.55 | 0.3 | |
| C3RL w/o RefBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 64.7 | 0.677 | 0.142 | |
| C3RLBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 64.5 | 0.696 | 0.107 | |
| RLCRBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 64.4 | 0.674 | 0.163 | |
| SCBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 59.2 | 0.744 | 0.146 | |
| SaySelfBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 56.6 | 0.761 | 0.117 | |
| BaseBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 54.9 | 0.554 | 0.337 | |
| SFT+RefBackbone Model=Qwen2.5VL-7B-Instruct2026.07 | 53.1 | 0.726 | 0.1 | |
| C3RLBackbone Model=Llama-3.2-3B-Instruct2026.07 | 50.1 | 0.691 | 0.159 | |
| C3RL w/o RefBackbone Model=Llama-3.2-3B-Instruct2026.07 | 49.3 | 0.679 | 0.147 | |
| RLVRBackbone Model=Llama-3.2-3B-Instruct2026.07 | 48.9 | 0.549 | 0.38 | |
| RLCRBackbone Model=Llama-3.2-3B-Instruct2026.07 | 46.7 | 0.709 | 0.112 | |
| BaseBackbone Model=Llama-3.2-3B-Instruct2026.07 | 43.9 | 0.569 | 0.276 | |
| SCBackbone Model=Llama-3.2-3B-Instruct2026.07 | 43.1 | 0.721 | 0.2 | |
| SFT+RefBackbone Model=Llama-3.2-3B-Instruct2026.07 | 40.8 | 0.598 | 0.19 | |
| SaySelfBackbone Model=Llama-3.2-3B-Instruct2026.07 | 23.3 | 0.667 | 0.081 |