Skip to content

bionemo-llm

The Bionemo Large Language Model (LLM) submodule contains common code used in submodules that train LLMs on biological datasets (currently bionemo-esm2 and bionemo-geneformer). This includes data masking and collate functions, the bio-BERT common architecture code, loss functions, and other NeMo / Megatron-LM compatibility functions. Sub-packages should only depend on bionemo-llm if they need access to NeMo and Megatron-LM.