List your company

Medical Question Answering on Medical QA Benchmarks (MedQA, MedMCQA, MMLU*, CMB, CMExam, CMMLU*) (test)

64.1MedQA AccuracyKaFT
48.333652.426856.5260.6132May 21, 2025

Evaluation results

MethodLinks
KaFTBackbone=LLaMA3-8B2025.0564.156.9474.0147.3947.8945.4455.961.8
No-conflictBackbone=LLaMA3-8B2025.0561.9856.1173.447.1747.7245.5955.331.17
Vanilla SFTBackbone=LLaMA3-8B2025.0561.8255.7573.3445.9945.8544.9554.620.46
Self-aligningBackbone=LLaMA3-8B2025.0561.3556.567347.5348.3446.5555.551.39
Vanilla SFTBackbone=Mistral-7B2025.0559.8643.7568.0636.2536.2535.4846.613.33
BaseBackbone=LLaMA3-8B2025.0559.6256.5172.6645.546.0444.6254.16—
KaFTBackbone=Mistral-7B2025.0559.5449.8768.4738.1137.3540.7349.015.73
No-conflictBackbone=Mistral-7B2025.0558.3751.1168.6938.0936.9236.1748.224.94
Self-aligningBackbone=Mistral-7B2025.0555.6250.9968.8136.736.8135.1347.344.06
Vanilla SFTBackbone=LLaMA3-3B2025.0554.9950.0862.943839.1738.1347.221.91
KaFTBackbone=LLaMA3-3B2025.0554.5250.5164.9340.1939.9339.748.32.99
KaFTBackbone=Qwen1.5-7B2025.0553.5749.8263.2375.5777.2772.0765.251.32
No-conflictBackbone=LLaMA3-3B2025.0552.9549.263.7638.9139.7837.2746.981.67
Self-aligningBackbone=LLaMA3-3B2025.0552.7949.8262.9338.5338.6638.5546.881.57
Vanilla SFTBackbone=Qwen1.5-7B2025.0552.7150.261.674.376.1570.1664.190.26
No-conflictBackbone=Qwen1.5-7B2025.0552.2450.5461.5575.0476.8770.7764.50.57
Self-aligningBackbone=Qwen1.5-7B2025.0551.7750.1163.1875.2377.0570.7164.670.74
BaseBackbone=LLaMA3-3B2025.0551.1449.4162.3435.9836.4836.545.31—
BaseBackbone=Mistral-7B2025.0550.9848.3165.0731.7330.6832.943.28—
BaseBackbone=Qwen1.5-7B2025.0548.9450.0862.7974.7776.5970.4263.93—