Paper gives KV cache eviction a probabilistic foundation

A UCLA paper reinterprets top-k KV cache eviction as probabilistic sampling estimation, providing a mathematical foundation and correction methods for a widespread yet theoretically unsupported LLM inference practice.

2026-09-01 ~ 2026-09-03 · 3 related posts

• Paper proposes probabilistic interpretation for KV Cache eviction with decode-time correction — burny_tech · 2026-09-01
• UCLA paper gives KV cache eviction a mathematical foundation via importance sampling — burkov · 2026-09-03

1 near-duplicate retellings: guyvdb