[None][perf] Avoid implicit device-scalar syncs in DeepSeek-V4 ctx sparse metadata#16734

qiaoxj07 merged 1 commit into

NVIDIA:mainNVIDIA/TensorRT-LLM:mainfrom

qiaoxj07:perf/dsv4-ctx-host-scalar-metadataqiaoxj07/TensorRT-LLM:perf/dsv4-ctx-host-scalar-metadataCopy head branch name to clipboard