Mathematical Reasoning on MATH500, AMC23, AIME24, AIME25
54.2Majority ScoreSDRL-freq
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SDRL-freqBackbone=Qwen3-4B-Base, Debate system agents (N)=5, Debate rounds=12026.01 | 54.2 | 57.7 | 3.5 | |
| SDRL-randBackbone=Qwen3-4B-Base, Debate system agents (N)=5, Debate rounds=12026.01 | 53.7 | 55.6 | 1.9 | |
| DAPOBackbone=Qwen3-4B-Base, Debate system agents (N)=5, Debate rounds=12026.01 | 52.8 | 52.9 | 0.1 | |
| SDRL-freqBackbone=Qwen2.5-3B, Debate system agents (N)=5, Debate rounds=12026.01 | 34.1 | 35.8 | 1.7 | |
| SDRL-randBackbone=Qwen2.5-3B, Debate system agents (N)=5, Debate rounds=12026.01 | 33.6 | 35.8 | 2.2 | |
| DAPOBackbone=Qwen2.5-3B, Debate system agents (N)=5, Debate rounds=12026.01 | 32.8 | 33.6 | 0.8 |