This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
Johannes "Lockhead" Koch
lockhead.dev
did:plc:yerl5oh6hes74chvhqumls23
Quadratic compute. On every single step. That's what LLM inference looks like without a KV cache.
The KV cache fixes it — making inference linear & affordable.
But your cluster might be destroying that cache without you even knowing. 😬
AWS User Group 🧵
2026-08-26T16:35:22.467Z