Comment by ryeguy

I keep seeing mention of the cache, what's special about it? All frontier llms have prefix caching, what is special about deepseek's approach?