Describir: Towards a Platform for Benchmarking Large Language Models