Mathematical Reasoning on AIME 2024 (Pass@1 and Pass@4)
66.87Pass@1AOPD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AOPDBackbone=Qwen3-8B-Base, SFT Warm-up steps=3K2026.05 | 66.87 | 79.58 | |
| GKDBackbone=Qwen3-8B-Base, SFT Warm-up steps=3K2026.05 | 66.25 | 80.65 | |
| ExOPDBackbone=Qwen3-8B-Base, SFT Warm-up steps=3K2026.05 | 61.67 | 77.55 | |
| OPDBackbone=Qwen3-8B-Base, SFT Warm-up steps=3K2026.05 | 61.46 | 78.61 | |
| AOPDBackbone=Qwen3-4B-Base, SFT Warm-up steps=3K2026.05 | 61.04 | 75.08 | |
| ExOPDBackbone=Qwen3-4B-Base, SFT Warm-up steps=3K2026.05 | 59.17 | 73.99 | |
| SFT CheckpointBackbone=Qwen3-8B-Base, SFT Warm-up steps=3K2026.05 | 58.96 | 74.95 | |
| GKDBackbone=Qwen3-4B-Base, SFT Warm-up steps=3K2026.05 | 58.13 | 76.61 | |
| OPDBackbone=Qwen3-4B-Base, SFT Warm-up steps=3K2026.05 | 57.08 | 76.69 | |
| GKDBackbone=Qwen3-8B-Base, SFT Warm-up steps=1K2026.05 | 54.79 | 77.92 | |
| AOPDBackbone=Qwen3-8B-Base, SFT Warm-up steps=1K2026.05 | 53.75 | 77.39 | |
| SFT CheckpointBackbone=Qwen3-4B-Base, SFT Warm-up steps=3K2026.05 | 52.29 | 71.9 | |
| SeqKDBackbone=Qwen3-8B-Base, SFT Warm-up steps=3K2026.05 | 51.04 | 76.56 | |
| ExOPDBackbone=Qwen3-8B-Base, SFT Warm-up steps=1K2026.05 | 45.83 | 72.31 | |
| SFT CheckpointBackbone=Qwen3-8B-Base, SFT Warm-up steps=1K2026.05 | 45.21 | 68.29 | |
| OPDBackbone=Qwen3-8B-Base, SFT Warm-up steps=1K2026.05 | 43.13 | 67.38 | |
| SeqKDBackbone=Qwen3-4B-Base, SFT Warm-up steps=3K2026.05 | 42.5 | 65.23 | |
| SeqKDBackbone=Qwen3-8B-Base, SFT Warm-up steps=1K2026.05 | 40.21 | 68.08 |