Enable Allgather/ReduceScatter backend for NaiveAllToAll (#23964)

Signed-off-by: Shu Wang. <shuw@nvidia.com>
Signed-off-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Signed-off-by: Shu Wang <shuw@nvidia.com>
Co-authored-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
Shu Wang
2025-09-18 15:52:58 +00:00
committed by GitHub
co-authored by Tyler Michael Smith Tyler Michael Smith Michael Goin
parent b419937c78
commit 2ea50e977a
3 changed files with 55 additions and 5 deletions
@@ -87,6 +87,10 @@ class CudaCommunicator(DeviceCommunicatorBase):
from .all2all import NaiveAll2AllManager
self.all2all_manager = NaiveAll2AllManager(self.cpu_group)
logger.info("Using naive all2all manager.")
elif all2all_backend == "allgather_reducescatter":
from .all2all import AgRsAll2AllManager
self.all2all_manager = AgRsAll2AllManager(self.cpu_group)
logger.info("Using AllGather-ReduceScatter all2all manager.")
elif all2all_backend == "pplx":
from .all2all import PPLXAll2AllManager
self.all2all_manager = PPLXAll2AllManager(self.cpu_group)