Medical Educational Explanation Generation on UK Medicine full per-architecture breakdown (Year 2)
0.0495BLEUILearner-LLM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| ILearner-LLMK-shot setting=52026.05 | 0.0495 | — | 0.7873 | 73.57 | 6.68 | 3.16 | |
| Gemma4-E4B-it SFTBackbone=Gemma4-E4B-it, Training Protocol=SFT2026.05 | 0.0476 | 0.8525 | 0.8372 | 67.86 | 16.04 | 3.0943 | |
| Qwen3-8B SFTBackbone=Qwen3-8B, Training Protocol=SFT2026.05 | 0.0399 | 0.8501 | 0.8352 | 64.3 | 16.32 | 2.49 | |
| RLearner-LLMBackbone=Gemma4-E4B, Training Protocol=RLearner-LLM2026.05 | 0.0325 | 0.8453 | 0.8397 | 66.55 | 38.92 | 3.0666 | |
| RLearner-LLMBackbone=Qwen3-8B, Training Protocol=RLearner-LLM2026.05 | 0.022 | 0.8129 | 0.796 | 79.14 | 20.09 | 3 | |
| RLearner-LLMBackbone=LLaMA-2-13B, Training Protocol=RLearner-LLM2026.05 | 0.0196 | 0.8247 | 0.8539 | 77.72 | 38.85 | 2.9738 | |
| SFTBackbone=LLaMA-2-13B, Training Protocol=SFT2026.05 | 0.0163 | 0.8208 | 0.8142 | 61.2 | 23.19 | 2.8717 |