Automatic evaluation of open-ended question answering in specialized domains remains challenging mainly because it relies on manual annotations from domain experts. In this work, we assess the ability of several large language models (LLMs), including closed-access (GPT-5.1, Gemini-2.5-Pro), open-source general-purpose (Qwen-80B), and biomedical domain-adapted models (MedGemma-27B, Phi-3.5-mini variants), to act as automatic evaluators of semantic equivalence in French medical open-ended QA. Our analysis reveals that LLM-based judgments are sensitive to the source of answer generation: judgement correlation varies substantially across different generator models. Among the judges, MedGemma-27B and Qwen-80B achieve the highest agreement with expert annotations in terms of F1 score and Pearson correlation. We further explore lightweight adaptation strategies on Phi-3.5-mini using supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO). Even with 184 training instances, these adaptations significantly improve Phi-3.5’s results and reduce variability across answer generators, achieving performance comparable to larger domain-adapted models. Our results highlight the importance of generator-aware evaluation, the limitations of general-purpose LLMs in domain-specific settings, and the effectiveness of lightweight adaptation for compact models in low-resource scenarios.

The development of large language models (LLMs) has led to increased focus on their adaptation to specialized domains and languages, yet the effectiveness of domain adaptation strategies remains unclear. We present a study of medical domain adaptation using French medical question answering (QA) as a case study. We compare continual pretraining (CPT), supervised fine-tuning (SFT), and their combination across three model families, multiple sizes, and three initialization types, explicitly disentangling adaptation effects from base model choice. We evaluate both multiple-choice (MCQA) and open-ended QA (OEQA) under greedy and constrained decoding using automatic metrics and LLM-as-a-Judge evaluation. For MCQA, CPT+SFT most often achieves the best scores, but gains over SFT are small and frequently not statistically significant, making SFT a strong and cost-effective default. For OEQA, CPT consistently improves overlap-based metrics, while SFT often degrades generation quality; instruction tuning and CPT+SFT are preferred by LLM-based evaluation. Cross-lingual experiments further show effective transfer from French adaptation to English benchmarks. Overall, we provide practical guidelines for selecting adaptation strategies under computational constraints.

Cet article présente une étude sur l’adaptation des grands modèles de langue (LLMs) à des domaines spécialisés disposant de données limitées. Bien que certaines recherches remettent en question le pré-entraînement adaptatif (DAPT) dans le contexte médical en anglais, nous montrons que l’adaptation au domaine peut être efficace sous certaines conditions. En prenant comme exemple l’adaptation au domaine médical en français, nous comparons de manière systématique le pré-entraînement continu (CPT), l’affinage supervisé (SFT) et une approche combinée (CPT suivi de SFT). Nos résultats indiquent que l’adaptation d’un modèle généraliste à de nouvelles données dans le domaine médical offre des améliorations notables (taux de réussite de 87%), tandis que l’adaptation supplémentaire de modèles déjà familiarisés avec ce domaine procure des bénéfices limités. Bien que CPT+SFT offre les meilleures performances globales, SFT-seul présente des résultats solides et requiert moins de ressources matérielles.