Domain Adaptation on MedQA, LegalBench, and FinQA
70.5MedQA ScoreFull Data
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Full DataModel=Qwen3-8B, Data Selection Ratio=100%2026.02 | 70.5 | 62.1 | 65.4 | 66 | |
| Full DataModel=Llama-3.1-8B, Data Selection Ratio=100%2026.02 | 65.2 | 58.4 | 61.3 | 61.6 | |
| OGSModel=Qwen3-8B, Data Selection Ratio=10%2026.02 | 64.1 | 59.5 | 62.8 | 62.1 | |
| LESSModel=Qwen3-8B, Data Selection Ratio=10%2026.02 | 62.3 | 55.4 | 59.2 | 59 | |
| GRADSModel=Qwen3-8B, Data Selection Ratio=10%2026.02 | 61.5 | 57.2 | 60.1 | 59.6 | |
| OGSModel=Llama-3.1-8B, Data Selection Ratio=10%2026.02 | 60.2 | 55.1 | 58.4 | 57.9 | |
| LESSModel=Llama-3.1-8B, Data Selection Ratio=10%2026.02 | 58.4 | 52.1 | 55.3 | 55.3 | |
| GRADSModel=Llama-3.1-8B, Data Selection Ratio=10%2026.02 | 57.9 | 53.5 | 56.1 | 55.8 | |
| Influence Func.Model=Qwen3-8B, Data Selection Ratio=10%2026.02 | 57.1 | 51.3 | 54.5 | 54.3 | |
| PerplexityModel=Qwen3-8B, Data Selection Ratio=10%2026.02 | 56.4 | 50.1 | 53.2 | 53.2 | |
| RandomModel=Qwen3-8B, Data Selection Ratio=10%2026.02 | 55.2 | 48.5 | 52.1 | 51.9 | |
| Influence Func.Model=Llama-3.1-8B, Data Selection Ratio=10%2026.02 | 53.1 | 47.2 | 50.1 | 50.1 | |
| PerplexityModel=Llama-3.1-8B, Data Selection Ratio=10%2026.02 | 52.3 | 46.8 | 49.5 | 49.5 | |
| RandomModel=Llama-3.1-8B, Data Selection Ratio=10%2026.02 | 50.1 | 45.3 | 48.2 | 47.9 |