Medical Question Answering on MedQA, MedMCQA, PubMedQA, MMLU-Pro-med, and GPQA-med (test)
57.61Accuracy (MedQA)For-Value
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| For-ValueBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=10%, Time=0.8 (h)2025.08 | 57.61 | 67.16 | 78.3 | 58.18 | 50.51 | 62.35 | |
| Hessian-freeBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=10%, Time=2.0 (h)2025.08 | 57.02 | 59.15 | 72.3 | 57.13 | 47.69 | 58.66 | |
| HyperINFBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=10%, Time=5.3 (h)2025.08 | 56.94 | 62.76 | 77.4 | 57.85 | 48.46 | 60.28 | |
| BaseBackbone=Llama-3.1-8B-Ins2025.08 | 56.84 | 61.9 | 77 | 59.02 | 44.35 | 59.82 | |
| For-ValueBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=5%, Time=0.8 (h)2025.08 | 56.8 | 62.92 | 77.6 | 58.31 | 45.9 | 60.31 | |
| DataInfBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=10%, Time=4.1 (h)2025.08 | 56.61 | 61.74 | 75.6 | 56.81 | 43.85 | 58.92 | |
| Hessian-freeBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=5%, Time=2.0 (h)2025.08 | 55.41 | 58.05 | 73.4 | 54.53 | 38.46 | 55.97 | |
| DataInfBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=5%, Time=4.1 (h)2025.08 | 55.39 | 57.74 | 73.3 | 54.07 | 45.13 | 57.13 | |
| HyperINFBackbone=Llama-3.1-8B-Ins, Data Selection Ratio=5%, Time=5.3 (h)2025.08 | 55.15 | 57.58 | 71.5 | 54.14 | 43.08 | 56.29 |