Medical Reasoning on MedBullets and MedXQA
84.87Accuracy (MedBullets)LLaMA3.2-3B-Instruct (Full-shot)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LLaMA3.2-3B-Instruct (Full-shot)Backbone Model=LLaMA3.2-3B-Instruct, Method Protocol=Full-shot, |Dtgt| (Target-domain data size)=1K, |Dgen| (General-domain data size)=02026.04 | 84.87 | 42.88 | 63.88 | |
| Qwen3-4B-Base (Full-shot)Backbone Model=Qwen3-4B-Base, Method Protocol=Full-shot, |Dtgt| (Target-domain data size)=1K, |Dgen| (General-domain data size)=02026.04 | 68.09 | 45.3 | 56.7 | |
| LLaMA3.2-3B-Instruct (HEAL)Backbone Model=LLaMA3.2-3B-Instruct, Method Protocol=HEAL, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=3842026.04 | 60.53 | 32.27 | 46.4 | |
| LLaMA3.2-3B-Instruct (Hybrid)Backbone Model=LLaMA3.2-3B-Instruct, Method Protocol=Hybrid, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=3842026.04 | 58.55 | 30.15 | 44.35 | |
| Qwen3-4B-Base (HEAL)Backbone Model=Qwen3-4B-Base, Method Protocol=HEAL, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=3842026.04 | 57.24 | 44.39 | 50.82 | |
| LLaMA3.2-3B-Instruct (Few-shot)Backbone Model=LLaMA3.2-3B-Instruct, Method Protocol=Few-shot, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=02026.04 | 56.91 | 31.36 | 44.14 | |
| Qwen3-4B-Base (Hybrid)Backbone Model=Qwen3-4B-Base, Method Protocol=Hybrid, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=3842026.04 | 55.92 | 42.73 | 49.33 | |
| Qwen3-4B-Base (Few-shot)Backbone Model=Qwen3-4B-Base, Method Protocol=Few-shot, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=02026.04 | 55.26 | 43.18 | 49.22 | |
| Qwen3-4B-Base (Only-General)Backbone Model=Qwen3-4B-Base, Method Protocol=Only-General, |Dtgt| (Target-domain data size)=0, |Dgen| (General-domain data size)=10K2026.04 | 49.67 | 39.7 | 44.69 | |
| Qwen3-1.7B-Base (HEAL)Backbone Model=Qwen3-1.7B-Base, Method Protocol=HEAL, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=3842026.04 | 49.34 | 40 | 44.67 | |
| Qwen3-1.7B-Base (Full-shot)Backbone Model=Qwen3-1.7B-Base, Method Protocol=Full-shot, |Dtgt| (Target-domain data size)=1K, |Dgen| (General-domain data size)=02026.04 | 49.01 | 40.45 | 44.73 | |
| LLaMA3.2-3B-Instruct (Only-General)Backbone Model=LLaMA3.2-3B-Instruct, Method Protocol=Only-General, |Dtgt| (Target-domain data size)=0, |Dgen| (General-domain data size)=10K2026.04 | 48.68 | 28.48 | 38.58 | |
| Qwen3-1.7B-Base (Hybrid)Backbone Model=Qwen3-1.7B-Base, Method Protocol=Hybrid, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=3842026.04 | 45.39 | 38.64 | 42.02 | |
| Qwen3-1.7B-Base (Few-shot)Backbone Model=Qwen3-1.7B-Base, Method Protocol=Few-shot, |Dtgt| (Target-domain data size)=32, |Dgen| (General-domain data size)=02026.04 | 45.07 | 38.33 | 41.7 | |
| Qwen3-4B-BaseBackbone Model=Qwen3-4B-Base, Method Protocol=Base2026.04 | 43.09 | 33.93 | 38.51 | |
| Qwen3-1.7B-Base (Only-General)Backbone Model=Qwen3-1.7B-Base, Method Protocol=Only-General, |Dtgt| (Target-domain data size)=0, |Dgen| (General-domain data size)=10K2026.04 | 42.11 | 32.73 | 37.42 | |
| Qwen3-1.7B-BaseBackbone Model=Qwen3-1.7B-Base, Method Protocol=Base2026.04 | 41.78 | 31.06 | 36.42 | |
| LLaMA3.2-3B-InstructBackbone Model=LLaMA3.2-3B-Instruct, Method Protocol=Base2026.04 | 40.46 | 26.82 | 33.64 |