2 papers
cs.AI2026
Runtime Observability for Heterogeneous Attention Memory
Fanzhe Wei, Li Liu, Ziyang Wang +1
Modern models no longer keep a plain KV cache: latent caches, learned sparse selectors and recurrent states each carry the model's memory in a different form, and each fails differ…
cs.AR2026
WitCert: Sound Runtime Risk Observability and Gating for KV-Cache Quantization
Fanzhe Wei, Li Liu, Ziyang Wang +1
KV-cache quantization is validated today by offline benchmark averages; a deployed system cannot tell whether compression is damaging the request it is serving right now. We give i…