Length Following on LIFEBench
10.8LD ScoreLARFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LARFTBase Model=Qwen2.5-7B-Instruct2026.02 | 10.8 | 80.57 | |
| LARFTBase Model=Llama-3.1-8B-Instruct2026.02 | 12.73 | 77.52 | |
| RLBase Model=Qwen2.5-7B-Instruct2026.02 | 17.17 | 70.94 | |
| LARFTBase Model=Llama-3.2-3B-Instruct2026.02 | 17.95 | 69.84 | |
| RLBase Model=Llama-3.1-8B-Instruct2026.02 | 19.74 | 67.39 | |
| LARFTBase Model=Qwen2.5-3B-Instruct2026.02 | 19.96 | 66.4 | |
| SFT+RLBase Model=Llama-3.2-3B-Instruct2026.02 | 20.41 | 66.48 | |
| SFT+RLBase Model=Qwen2.5-7B-Instruct2026.02 | 21.42 | 65.15 | |
| SFT+RLBase Model=Llama-3.1-8B-Instruct2026.02 | 21.43 | 65.14 | |
| RLBase Model=Llama-3.2-3B-Instruct2026.02 | 22.3 | 64.02 | |
| RLBase Model=Qwen2.5-3B-Instruct2026.02 | 30.71 | 54.1 | |
| OriginBase Model=Qwen2.5-7B-Instruct2026.02 | 30.96 | 53.84 | |
| SFT+RLBase Model=Qwen2.5-3B-Instruct2026.02 | 32.48 | 52.23 | |
| OriginBase Model=Qwen2.5-3B-Instruct2026.02 | 38.79 | 46.04 | |
| OriginBase Model=Llama-3.1-8B-Instruct2026.02 | 41.13 | 43.93 | |
| OriginBase Model=Llama-3.2-3B-Instruct2026.02 | 44.66 | 40.93 | |
| SFTBase Model=Llama-3.1-8B-Instruct2026.02 | 44.96 | 42.64 | |
| PositionIDBase Model=Llama-3.1-8B-Instruct2026.02 | 53.95 | 33.99 | |
| RulerBase Model=Llama-3.1-8B-Instruct2026.02 | 54.65 | 33.52 | |
| RulerBase Model=Qwen2.5-7B-Instruct2026.02 | 57.22 | 31.84 | |
| PositionIDBase Model=Qwen2.5-7B-Instruct2026.02 | 57.67 | 31.56 | |
| SFTBase Model=Qwen2.5-7B-Instruct2026.02 | 60.12 | 30.05 | |
| SFTBase Model=Llama-3.2-3B-Instruct2026.02 | 61.88 | 29.01 | |
| RulerBase Model=Llama-3.2-3B-Instruct2026.02 | 62.11 | 28.87 | |
| RulerBase Model=Qwen2.5-3B-Instruct2026.02 | 62.15 | 28.85 | |
| PositionIDBase Model=Llama-3.2-3B-Instruct2026.02 | 63.57 | 28.04 | |
| PositionIDBase Model=Qwen2.5-3B-Instruct2026.02 | 64.95 | 27.28 | |
| SFTBase Model=Qwen2.5-3B-Instruct2026.02 | 65.6 | 26.93 |