JPharmatron & JPharmaBench: A Japanese LLM and Evaluation Benchmarks for Pharmaceutical NLP

Jan 13, 2026
Code Paper View Slides

Overview

This talk was given at the 25th LLM-jp勉強会 (LLM-jp Seminar) on January 13, 2026. I presented our work on JPharmatron, the first Japanese language model specialized for pharmaceutical operations, and JPharmaBench, a benchmark suite consisting of three novel evaluation benchmarks for pharmaceutical NLP:

  • YakugakuQA: 3,021 questions from the Japanese National License Examination for Pharmacists
  • NayoseQA: 34,769 cross-lingual drug/chemical name normalization questions from KEGG DRUG
  • SogoCheck: 200 cross-document inconsistency detection tasks from the Japanese Pharmacopoeia

The presentation was given in Japanese.

AI LLM Language Models NLP Presentations
Shinnosuke Ono
Shinnosuke Ono
Master’s Student

I’m interested in understanding how machines learn through the lens of representations. My research interests include representation learning, multimodal models, language models, and reinforcement learning.