[None][fix] Derive KV capacity once and select windowed blocks by ordinal - #18668

[None][fix] Derive KV capacity once and select windowed blocks by ordinal#18668

Shixiaowei02 wants to merge 1 commit into

NVIDIA:mainNVIDIA/TensorRT-LLM:mainfrom

Shixiaowei02:user/xiaoweis/kv-capacity-contractShixiaowei02/TensorRT-LLM:user/xiaoweis/kv-capacity-contractCopy head branch name to clipboard