Repository navigation

feat(kvcache): host-RAM KV tier for sparse-attention models (--kv-host-pages) - #499

Open

alvarorsouza-arch wants to merge 3 commits into

FlashML-org:mainfrom

alvarorsouza-arch:qsa-kv-host-offload

Commits on Sep 17, 2026

feat(kvcache): host-RAM KV tier for sparse-attention models (--kv-host-pages)
alvarorsouza-arch
committed
849e072

Commits on Sep 19, 2026

fix(kvcache): fp8-aware write-through mirror + batch-safe selection cap
alvarorsouza-arch
committed
20f3b23
fix(qsa): 16-bit compute dtype guard for the indexer + fixed-shape write-pages
alvarorsouza-arch
committed
6c316e2