large language models (LLMs) have been driving progress in AI at an
unprecedented rate, yet still face challenges in knowledge-intensive domains
like biomedicine. Solutions such as pre-training and domain-specific
fine-tuning add substantial computational overhead, and the latter requi