I think it is a great start, and I suppose there will be many iterations to ensure fair usage. It would be interesting to reach a point to be similar to docker

I think it is a great start, and I suppose there will be many iterations to ensure fair usage. It would be interesting to reach a point to be similar to docker where you don't need to load each layer again and you only need your specific layer. The shared models layers would be already loaded, and running multiple models at once would consume less GPU memory.