List your company

Medical Question Answering on Out-of-domain medical QA History, Engineering, Law (test)

50.4History AccuracyKaFT
32.92837.4644246.536May 21, 2025

Evaluation results

MethodLinks
KaFTBackbone=Mistral-7B2025.0550.46051.6546.93
Vanilla SFTBackbone=Mistral-7B2025.054659.25051.734.66
No-conflictBackbone=Mistral-7B2025.0545.261.249.6524.93
Self-aligningBackbone=Mistral-7B2025.0544.456.849.250.133.06
BaseBackbone=Mistral-7B2025.0541.253.246.847.07—
No-conflictBackbone=LLAMA3-3B2025.0540.8584848.936.66
KaFTBackbone=LLAMA3-3B2025.0540.458.449.649.477.2
Vanilla SFTBackbone=LLAMA3-3B2025.054056.446.447.65.33
Self-aligningBackbone=LLAMA3-3B2025.0539.656.448485.73
BaseBackbone=LLAMA3-3B2025.0533.654.838.442.27—