In order to ensure that we do not have regression in the compose time of models, we should have tests for a variety of model sizes. Ideally this would integrate with BenchmarkDotNet to make sure we are not adding significant runtime or memory allocations.