Legal Reasoning and Text Generation on Auckland Law
13.82BLEUQwen3-8B SFT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen3-8B SFTBackbone=Qwen3-8B2026.05 | 13.82 | 87.84 | 85.57 | 51.75 | 31.91 | 2 | |
| Gemma4-E4B-it SFTBackbone=Gemma4-E4B-it2026.05 | 13.53 | 87.98 | 86.11 | 51.72 | 39.11 | 2.6431 | |
| RLearner-LLMBackbone=Gemma4-E4B2026.05 | 13.15 | 87.57 | 86.54 | 55.63 | 43.77 | 2.6304 | |
| RLearner-LLMBackbone=LLaMA-22026.05 | 4.57 | 82.65 | 82.97 | 55.46 | 32.29 | 2.5918 | |
| ILearner-LLMIteration Count (K)=52026.05 | 3.81 | — | 77.2 | 63.26 | 39.96 | 2.7845 | |
| RLearner-LLMBackbone=Qwen32026.05 | 3.62 | 81.46 | 80.05 | 80.7 | 23.03 | 2.67 | |
| SFTBackbone=LLaMA-2-13B2026.05 | 2.98 | 80.1 | 77.09 | 55.16 | 27.02 | 2.718 |