Repository navigation

perf(phi4_multimodal): prefill and decode over the cached prefix, not the whole KV allocation - #2202

Merged

chaofengw-nv merged 2 commits into

NVIDIA:mainfrom

chaofengw-nv:perf/phi4mm-prefill-cache-rows

Nov 11, 2026

Merged

perf(phi4_multimodal): prefill and decode over the cached prefix, not the whole KV allocation#2202
chaofengw-nv merged 2 commits into
NVIDIA:mainfrom
chaofengw-nv:perf/phi4mm-prefill-cache-rows

Commits on Nov 10, 2026
- perf(phi4_multimodal): prefill over the cached prefix, not the whole KV allocation
  chaofengw-nv
  committed
  4d72959
- perf(phi4_multimodal): decode over the cached prefix, not the whole KV allocation
  chaofengw-nv
  committed
  d8d899f