TheBioCollection-Eval is a biological evaluation suite for assessing large language models (BioLMs) for biology across small molecules, proteins, genomic sequences, cells/pathways, and cross-domain reasoning. It is constructed by drawing subtasks from many scattered existing benchmarks (Mol-Instructions, MolLangBench, BioReason-Pro, PerturBench Replogle K562) and combining them with source-derived newly-constructed instruction datasets.
