I think there is a lot of value in people documenting what it takes to make models that will run on 8, 12, and 16GB GPUs work well for them, and the tools used.