Mathematical Reasoning on AIME 25 (Acc, Expl., Fmt)
16.7AccuracyLACE: + SFT & RL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LACE: + SFT & RLModel Scale=Qwen3-4B2026.04 | 16.7 | 14 | 100 | |
| LACE: + SFT & RLModel Scale=Qwen3-1.7B2026.04 | 13.3 | 31 | 90 | |
| Independent Sampling: Standard SFT + RLModel Scale=Qwen3-4B2026.04 | 13.3 | 7 | — | |
| Independent Sampling: + VotingModel Scale=Qwen3-4B2026.04 | 13.3 | 7 | — | |
| Independent Sampling: + VotingModel Scale=Qwen3-1.7B2026.04 | 10 | 26 | — | |
| LACE: + SFTModel Scale=Qwen3-1.7B2026.04 | 10 | 14 | 40 | |
| Independent Sampling: Standard SFT + RLModel Scale=Qwen3-1.7B2026.04 | 8.3 | 26 | — | |
| LACE: Continuous Pre-trainModel Scale=Qwen3-1.7B2026.04 | 6.7 | 41 | 50 | |
| Isolated Parallel: + SFTModel Scale=Qwen3-4B2026.04 | 6.7 | 11 | 23.3 | |
| Isolated Parallel: + SFT & RLModel Scale=Qwen3-4B2026.04 | 6.7 | 8 | 86.7 | |
| Isolated Parallel: + SFT & RLModel Scale=Qwen3-1.7B2026.04 | 3.3 | 18 | 53.3 | |
| Isolated Parallel: + VotingModel Scale=Qwen3-4B2026.04 | 3.3 | 8 | 86.7 | |
| LACE: + SFTModel Scale=Qwen3-4B2026.04 | 3.3 | 12 | 40 | |
| Isolated Parallel: + SFTModel Scale=Qwen3-1.7B2026.04 | 0 | 22 | 46.7 | |
| Isolated Parallel: + VotingModel Scale=Qwen3-1.7B2026.04 | 0 | 18 | 53.3 | |
| LACE: Continuous Pre-trainModel Scale=Qwen3-4B2026.04 | 0 | 22 | 0 |