Mathematical Reasoning on Olympiad (Acc, AUROC, ECE)
17AccuracyRLVR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| RLVRBackbone=Llama-3.2-3B-Instruct2026.07 | 17 | 0.593 | 0.754 | |
| C3RL w/o RefBackbone=Llama-3.2-3B-Instruct2026.07 | 16.2 | 0.656 | 0.278 | |
| C3RL (Ours)Backbone=Llama-3.2-3B-Instruct2026.07 | 15.8 | 0.641 | 0.289 | |
| RLCRBackbone=Llama-3.2-3B-Instruct2026.07 | 12.8 | 0.657 | 0.077 | |
| SCBackbone=Llama-3.2-3B-Instruct2026.07 | 8.6 | 0.631 | 0.12 | |
| BaseBackbone=Llama-3.2-3B-Instruct2026.07 | 8.3 | 0.559 | 0.379 | |
| SFT+RefBackbone=Llama-3.2-3B-Instruct2026.07 | 5.2 | 0.583 | 0.17 | |
| SaySelfBackbone=Llama-3.2-3B-Instruct2026.07 | 3.3 | 0.654 | 0.083 |