To reduce the memory need of our API service, we can delegate stabilization to dedicated workers. Without this, the API needs to have enough memory to load the entire artifact. That memory is going unused for the majority of an API call's lifetime while it waits for inference and build execution to complete.