[Elastic EP] Async preparation (#47288)
Signed-off-by: Itay Alroy <ialroy@nvidia.com>
This commit is contained in:
@@ -36,6 +36,7 @@ class CudaCommunicator(DeviceCommunicatorBase):
|
||||
global_ranks: list[int] | None = None,
|
||||
global_world_size: int | None = None,
|
||||
tcp_store_group: StatelessProcessGroup | None = None,
|
||||
use_all2all: bool = False,
|
||||
):
|
||||
super().__init__(
|
||||
cpu_group,
|
||||
@@ -44,6 +45,7 @@ class CudaCommunicator(DeviceCommunicatorBase):
|
||||
unique_name,
|
||||
global_ranks,
|
||||
global_world_size,
|
||||
use_all2all=use_all2all,
|
||||
)
|
||||
if "tp" not in unique_name:
|
||||
# custom allreduce or torch symm mem can be used only by tp
|
||||
|
||||
Reference in New Issue
Block a user