[MoE Refactor] Remove MoE DP chunking (#39107)
Signed-off-by: Bill Nell <bnell@redhat.com> Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
This commit is contained in:
@@ -622,6 +622,18 @@ class ParallelConfig:
|
||||
and self.data_parallel_size > 1
|
||||
)
|
||||
|
||||
@property
|
||||
def use_batched_dp_moe(self) -> bool:
|
||||
return (
|
||||
self.all2all_backend
|
||||
in (
|
||||
"deepep_low_latency",
|
||||
"nixl_ep",
|
||||
)
|
||||
and self.enable_expert_parallel
|
||||
and self.data_parallel_size > 1
|
||||
)
|
||||
|
||||
@property
|
||||
def node_rank_within_dp(self) -> int:
|
||||
return self.node_rank % self.nnodes_within_dp
|
||||
|
||||
Reference in New Issue
Block a user