How do caches work across models? I would have thought that was very model specific - if not I’ve really misunderstood what’s getting cached.