ParentFull threadjbellis·Deepseek pioneered automatic prefix caching and caches on SSD. SSD reads are so fast compared to LLM inference that I can't think of a reason to waste ram on it.View on HN