From 300KB to 69KB per Token: How LLM Architectures Solve the KV Cache Problemnews.future-shock.ai
74 points by future-shock-ai 3 days ago | 5 comments