For effective analysis of model performance, benchmarks are needed. This task includes doing a thorough literature review, and seeing what benchmarks seem likely to be useful for downstream.