From 300KB to 69KB per Token: How LLM Architectures Solve the KV Cache Problem 39 points by future-shock-ai 2 days ago 5 comments story