BMRetriever: Tuning Large Language Models as Better Biomedical Text Retrievers
This work addresses the problem of biomedical retrieval for researchers and practitioners by providing efficient models that reduce computational resource needs, though it is incremental as it builds on existing dense retrieval methods.
The authors tackled the challenge of biomedical text retrieval by developing BMRetriever, a series of dense retrievers that use unsupervised pre-training and instruction fine-tuning, achieving strong performance with the 410M variant outperforming baselines up to 11.7 times larger and the 2B variant matching models with over 5B parameters.
Developing effective biomedical retrieval models is important for excelling at knowledge-intensive biomedical tasks but still challenging due to the deficiency of sufficient publicly annotated biomedical data and computational resources. We present BMRetriever, a series of dense retrievers for enhancing biomedical retrieval via unsupervised pre-training on large biomedical corpora, followed by instruction fine-tuning on a combination of labeled datasets and synthetic pairs. Experiments on 5 biomedical tasks across 11 datasets verify BMRetriever's efficacy on various biomedical applications. BMRetriever also exhibits strong parameter efficiency, with the 410M variant outperforming baselines up to 11.7 times larger, and the 2B variant matching the performance of models with over 5B parameters. The training data and model checkpoints are released at \url{https://huggingface.co/BMRetriever} to ensure transparency, reproducibility, and application to new domains.