I am working on a benchmark and need to manage several interlocking components: datasets and metadata, diverse ML tasks (varying inputs and outputs), and baseline experiments covering models, training, and evaluations. Any pointers to projects that handle these through clean/minimal data structures
