#kv-cache

The Inference Machine: KV Cache, Attention, and How Modern LLMs Stay Fast