From the 1 of 3 linked papers with an AI index.
3 papers
cs.DC2026
Route-Block Membership Selects Packed-AWQ Arithmetic: A Controlled Single-Fixture Mechanism Study
Lukas Stepanek
The paper studies how the alignment of routed tokens into expert blocks determines the exact packed quantized matrix multiplication performed during Mixture‑of‑Experts inference, d…
cs.DC2026
Fail-Closed Lowering of Resident KV Claims onto LLM Serving Runtimes
Lukas Stepanek
LLM serving runtimes increasingly expose KV-cache primitives that resemble future-reuse controls: retention priority, TTL-like duration, host or storage offload, block events, acti…
cs.DC2026
Resident KV Claims: A Conformance Contract for Future Reuse under Active KV Pressure
Lukas Stepanek
KV-cache reuse mechanisms increasingly expose priority, duration, offload, routing hints, scheduler modes, and event streams. These mechanisms help preserve reusable prefixes, but…