CIMILab is excited to announce the launch of BioMed-LLM, our new research project on domain-adapted large language models for biomedical applications. Led by Carlos Mendez (MSc student) under the supervision of Dr. Dey, the project will explore fine-tuning and retrieval-augmented generation (RAG) strategies for open-source LLMs on clinical NLP tasks.
Initial benchmarks on PubMedQA and MedQA show promising results, with our fine-tuned models outperforming zero-shot GPT-3.5 baselines. We'll be releasing code and model checkpoints progressively — follow the GitHub repository for updates.