points
But if not, does anybody know a recommended way to attach vision to deepseek flash (on a self-hosted infrastructure)?