Mathematical Reasoning on LIMO
17,474LengthFull CoT SFT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Full CoT SFTDecoding Strategy=Greedy2026.01 | 17,474 | 0.304 | — | |
| Segment Selective SFTDecoding Strategy=Greedy2026.01 | 16,005 | 0.338 | — | |
| Full CoT SFTDecoding Strategy=Sampling2026.01 | 15,067 | — | 33 | |
| Segment Selective SFTDecoding Strategy=Sampling2026.01 | 14,669 | — | 33.5 | |
| LLaMA3.1-8B-InstructDecoding Strategy=Greedy2026.01 | 9,691 | 0.24 | — | |
| LLaMA3.1-8B-InstructDecoding Strategy=Sampling2026.01 | 4,161 | — | 23.5 |