Papers /
Small language models learn enhanced reasoning skills from medical textbooks
Hyunjae Kim, Hyeon Hwang, Jiwoo Lee, Jaewoo Kang et al. · npj Digital Medicine · 2025-05-02 · doi · code
What it does (our summary of the abstract)
Fine-tunes small open language models (7B and 8B parameters) on step-by-step reasoning drawn from medical textbooks, so that they answer medical exam questions by reasoning before choosing an option.
Model at a glance (from the paper's methods)
| Input | Medical exam question text (English training data) |
|---|---|
| Output | Free-text reasoning and an answer choice |
| Data | Chain-of-thought data built from medical textbooks and exam questions |
| Weights | Released on Hugging Face (Meerkat-7B, Meerkat-8B) |
Validated on SNUH data
clinicalqa@0 preview · ClinicalQA: Korean Medical Knowledge QA Benchmark, 1045 test instances
accuracy 0.674 (0.645–0.700) · #17 of 17 · agent (run by the medbench paper agent with the authors' released weights) · leaderboard · result
How the agent ran it: ✓ weights released 0.44 GPU-h
Authors: submit your own run to get an official entry on this task. See For agents.