Question Answering on expert-curated (test)
31.65Token F1DoRA SFT with human-annotated supervision
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DoRA SFT with human-annotated supervisionRetrieval mechanism=GTE-based, Supervision strategy=Human-annotated2026.04 | 31.65 | 29.28 | 6.78 | -0.4 | 75.88 | |
| Llama3.1-8B-Instruct (base)Retrieval mechanism=GTE-based2026.04 | 25.27 | 23.5 | 6.62 | -0.827 | 70.98 | |
| GPT-4oRetrieval mechanism=GTE-based2026.04 | 25.19 | 23.61 | 5.9 | -0.793 | 71.05 | |
| DoRA SFTRetrieval mechanism=GTE-based, Supervision strategy=Synthetic-only2026.04 | 22.96 | 23.07 | 2 | -0.565 | 72.26 |