Exploring Post-Training Alignment of Small Language Models for Biomedical Data-to-Text Generation: A Case Study of Medication Leaflet

2026-07-15

AI Summary

This study addresses the critical challenge in biomedical informatics of translating complex biomedical data into patient-comprehensible text, focusing specifically on drug label generation. It presents the first systematic evaluation of alignment techniques—including supervised fine-tuning (SFT), direct preference optimization (DPO), odds ratio preference optimization (ORPO), and group relative policy optimization (GRPO)—applied to the Qwen small language model. Comprehensive multidimensional assessment using ROUGE metrics and semantic similarity on a newly constructed dataset combining drug labels and openFDA data demonstrates that aligned small models consistently outperform GPT-5. Among the methods, ORPO significantly surpasses SFT, while GRPO exhibits the most robust generalization across datasets. These findings establish an efficient and scalable alignment paradigm for biomedical text generation.

Abstract

Translating complex biomedical data into patient-friendly narratives is central to modern biomedical informatics. This study presents a comparative analysis of training small language models (SLMs) in specialized biomedical datato-text generation tasks. We explore widely adopted post-training methods including supervised fine-tuning (SFT), direct preference optimization (DPO), odds ratio preference optimization (ORPO), and group relative policy optimization (GRPO) with Qwen-based SLMs on a medicine package leaflets dataset. To assess cross-dataset generalizability, we also curated drug label data from openFDA. We evaluate models using both standard lexical overlap metrics like ROUGE as well as semantic similarity measures. Across our experiments, the results show that (1) the aligned SLMs outperform proprietary models like GPT-5; (2) ORPO outperforms the SFTbaselines; (3) GRPO yields the most robust cross-dataset performance among the alignment methods tested as well as GPT-5.

Problem

• biomedical data-to-text generation
• small language models
• post-training alignment
• medication leaflet
• patient-friendly narratives

Innovation

• post-training alignment
• small language models
• biomedical data-to-text generation
• ORPO
• GRPO