Ollama has a way to override where u the models get downloaded. so u essentially create a network volume on serverless under /runpod-volume is where they get mounted for serverless
And when ur ollama server starts through a background script on start, u do whatever u want. overall its a bit of a pain