Scientific Reasoning on Sydney Biology per-architecture breakdown (full)
8.93BLEURLearner-LLM
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RLearner-LLMBackbone=Gemma4-E4B2026.05 | 8.93 | 88.03 | 84.72 | 54.21 | 23.09 | 3.0142 | 4.667 | |
| Qwen3-8B SFTBackbone=Qwen3-8B2026.05 | 8.44 | 87.88 | 84.16 | 59.29 | 17.37 | 2.26 | — | |
| Gemma4-E4B-it SFTBackbone=Gemma4-E4B-it2026.05 | 8.44 | 87.94 | 84.52 | 62.09 | 24.69 | 3.0364 | 5.804 | |
| RLearner-LLMBackbone=Qwen32026.05 | 4.18 | 83.53 | 79.95 | 82.83 | 22.84 | 2.78 | — | |
| DPO v2Backbone=LLaMA-2-13B, Note=ours2026.05 | 3.64 | 83.67 | 84.26 | 62.9 | 27.74 | 2.9474 | 6.37 | |
| RLearner-LLMBackbone=LLaMA-22026.05 | 3.16 | 83.59 | 86.2 | 63.27 | 35.62 | 2.8376 | 4.06 | |
| ILearner-LLMK=52026.05 | 2.74 | — | 78.89 | 64.51 | 8.37 | 3.1843 | — | |
| SFTBackbone=LLaMA-2-13B2026.05 | 2.22 | 82.44 | 78.7 | 62.49 | 5.37 | 3.1937 | 19.001 |