Mathematical Reasoning on AMOBench (Pass@16)
61.5Pass@16TRD
Evaluation Results
| Method | Links | |
|---|---|---|
| TRDBackbone=Qwen3-8B (w/ thinking), Trajectory=yr2026.06 | 61.5 | |
| Forward KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 51.3 | |
| Reverse KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 51.3 | |
| Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 43.6 | |
| BaseBackbone=Qwen3-8B (w/ thinking), Trajectory=–2026.06 | 41 | |
| Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.06 | 38.5 | |
| TRDModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yr, Teacher Model=Qwen3-8B2026.06 | 35.9 | |
| TRDBackbone=Qwen3-4B-Instruct-2507, Trajectory=yr2026.06 | 33.8 | |
| Forward KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 33.3 | |
| Forward KLModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 33.3 | |
| Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 28.2 | |
| Reverse KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 28.2 | |
| Reverse KL w/ Top-KModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 28.2 | |
| Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.06 | 25.6 | |
| Forward KL w/ ClipModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 25.6 | |
| BaseBackbone=Qwen3-4B-Instruct-2507, Trajectory=–2026.06 | 23.1 | |
| BaseModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=-, Teacher Model=Qwen3-8B2026.06 | 23.1 | |
| TRDModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yr, Teacher Model=Qwen3-8B2026.06 | 17.9 | |
| Reverse KLModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 17.9 | |
| Reverse KL w/ Top-KModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 15.4 | |
| BaseModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=-, Teacher Model=Qwen3-8B2026.06 | 12.8 | |
| Forward KLModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 12.8 | |
| Reverse KLModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 12.8 | |
| Forward KL w/ ClipModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.06 | 10.3 |