Skip to content

cuda : reduce large tensors through the copy engines when NCCL has no peer access - #16

Merged
Piggidragon merged 2 commits into
mainfrom
cuda/allreduce-copy-engine
Oct 16, 2026
Merged

cuda : reduce large tensors through the copy engines when NCCL has no peer access#16
Piggidragon merged 2 commits into
mainfrom
cuda/allreduce-copy-engine

Commits

  1. Commits on Oct 16, 2026