动手理解 AI 的可交互教学博客https://lab.door4ai.com/https://lab.door4ai.com/posts/kv-cache-quantization/https://lab.door4ai.com/posts/kv-cache-quantization/从一组数怎么压成 4 bit，讲到 DeepSeek 为什么能把 KV cache 砍掉 90%。五个可交互实验。Tue, 13 Oct 2026 00:00:00 GMT