microsoft/biogpt

BioGPT is a language model from Microsoft Research trained specifically on biomedical text. While general-purpose language models are trained on broad internet text, BioGPT was pre-trained on a large collection of biomedical research papers from PubMed, making it better suited for tasks involving medical and scientific language. The repository provides code and pre-trained model weights for several versions of the model, including a larger variant called BioGPT-Large. Beyond the base pre-trained model, fine-tuned versions are also available for specific tasks: extracting relationships between biological entities (such as drug-drug interactions or drug-target interactions), classifying documents by topic, answering questions from biomedical literature, and generating biomedical text. There are two main ways to use the models. The original method uses the fairseq training framework, which requires several additional tools installed from source, including a tokenizer and a byte-pair encoding tool. The simpler method is through the Hugging Face transformers library, where the model is available as a standard component and can be loaded with just a few lines of Python code. The Hugging Face route is substantially easier for most users. The models run on GPU and are intended for researchers and developers working in biomedical natural language processing. Practical use cases mentioned in the repository include mining scientific literature for entity relationships, building question-answering systems over medical papers, and generating biomedical text that continues a given prompt. This is a research release. It is not a consumer product or clinical tool, and the README does not describe any production deployment pathway.