Reasoning on LiveBench (Acc, Expl., Fmt)
33AccuracyLACE: + SFT & RL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LACE: + SFT & RLModel Scale=Qwen3-4B2026.04 | 33 | 13 | 85.5 | |
| Independent Sampling: + VotingModel Scale=Qwen3-4B2026.04 | 28 | 8 | — | |
| Independent Sampling: Standard SFT + RLModel Scale=Qwen3-4B2026.04 | 25.6 | 8 | — | |
| LACE: + SFT & RLModel Scale=Qwen3-1.7B2026.04 | 16.5 | 23 | 53.5 | |
| Independent Sampling: + VotingModel Scale=Qwen3-1.7B2026.04 | 16 | 20 | — | |
| LACE: + SFTModel Scale=Qwen3-4B2026.04 | 16 | 10 | 31 | |
| Isolated Parallel: + VotingModel Scale=Qwen3-1.7B2026.04 | 15.5 | 26 | 57.5 | |
| Isolated Parallel: + VotingModel Scale=Qwen3-4B2026.04 | 15.5 | 7 | 45.5 | |
| Independent Sampling: Standard SFT + RLModel Scale=Qwen3-1.7B2026.04 | 14.6 | 20 | — | |
| Isolated Parallel: + SFT & RLModel Scale=Qwen3-1.7B2026.04 | 14 | 26 | 57.5 | |
| Isolated Parallel: + SFT & RLModel Scale=Qwen3-4B2026.04 | 11 | 7 | 49 | |
| LACE: + SFTModel Scale=Qwen3-1.7B2026.04 | 7.5 | 23 | 32.5 | |
| Isolated Parallel: + SFTModel Scale=Qwen3-4B2026.04 | 7 | 10 | 29.5 | |
| LACE: Continuous Pre-trainModel Scale=Qwen3-1.7B2026.04 | 5 | 23 | 22 | |
| Isolated Parallel: + SFTModel Scale=Qwen3-1.7B2026.04 | 3.5 | 25 | 25 | |
| LACE: Continuous Pre-trainModel Scale=Qwen3-4B2026.04 | 1 | 25 | 2 |