Repository navigation

perf(data-plane): avoid local batch copies - #4091

Closed

rohitrango wants to merge 5 commits into

NVIDIA-NeMo:rohit/packedtensor_paddingfrom

rohitrango:rohit/packedtensor_opt

Closed

perf(data-plane): avoid local batch copies#4091
rohitrango wants to merge 5 commits into
NVIDIA-NeMo:rohit/packedtensor_paddingfrom
rohitrango:rohit/packedtensor_opt

Commits on Sep 10, 2026
- perf(data-plane): avoid local batch copies
  rohitrango
  committed
  33d6b8e
- perf(data-plane): preserve PackedTensor dedup on broadcast
  rohitrango
  committed
  b7663d1
- refactor(data-plane): derive PackedTensor broadcast state
  rohitrango
  committed
  696bd95
Commits on Sep 11, 2026
- fix(data-plane): bound PackedTensor broadcast memory
  rohitrango
  committed
  a609bff
- perf(data-plane): coalesce packed payloads on GPU
  rohitrango
  committed
  0d622ce