Math on HMMT25 (Avg@16)
44.5Avg@16TRD
Evaluation Results
| Method | Links | |
|---|---|---|
| TRDBackbone=Qwen3-8B (w/ thinking), Traj.=yr2026.06 | 44.5 | |
| +Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 44.4 | |
| +Reverse KLBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 44.3 | |
| +Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 42.5 | |
| BaseBackbone=Qwen3-8B (w/ thinking), Traj.=–2026.06 | 41.5 | |
| +Forward KLBackbone=Qwen3-8B (w/ thinking), Traj.=yo2026.06 | 40.3 | |
| TRDBackbone=Qwen3-4B-Instruct-2507, Traj.=yr2026.06 | 32.1 | |
| BaseBackbone=Qwen3-4B-Instruct-2507, Traj.=–2026.06 | 31.3 | |
| +Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 31 | |
| +Forward KLBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 30.8 | |
| +Reverse KLBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 30 | |
| +Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Traj.=yo2026.06 | 27.9 |