Math on BeyondAIME (Avg@16)
42.8Avg@16TRD
Evaluation Results
| Method | Links | |
|---|---|---|
| TRDBackbone=Qwen3-8B (w/ thinking), Traj.=yr2026.06 | 42.8 | |
| +Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 41.9 | |
| BaseBackbone=Qwen3-8B (w/ thinking), Traj.=–2026.06 | 41.6 | |
| +Reverse KLBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 40.8 | |
| +Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 40.2 | |
| +Forward KLBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 39.6 | |
| TRDBackbone=Qwen3-4B-Instruct-2507, Traj.=yr2026.06 | 32.7 | |
| +Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 32.4 | |
| +Forward KLBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 32.3 | |
| +Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 32.3 | |
| BaseBackbone=Qwen3-4B-Instruct-2507, Traj.=–2026.06 | 32.2 | |
| +Reverse KLBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 31.6 |