This story has moved to /story/indexmem-learned-kv-cache-eviction-with-latent-memory-for-long-context-llm-infer/.