[MoE Refactor] EPLB refactoring for FusedMoE (#41055)

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:
bnellnm
2026-05-12 14:16:31 -04:00
committed by GitHub
co-authored by Robert Shaw
parent 4d591db470
commit d9b4990783
17 changed files with 83 additions and 98 deletions
+11
View File
@@ -940,6 +940,17 @@ class EplbLayerState:
GPU work.
"""
def set_layer_state(
self,
moe_layer_idx: int,
expert_load_view: torch.Tensor,
logical_to_physical_map: torch.Tensor,
logical_replica_count: torch.Tensor,
) -> None:
self.expert_load_view = expert_load_view[moe_layer_idx]
self.logical_to_physical_map = logical_to_physical_map[moe_layer_idx]
self.logical_replica_count = logical_replica_count[moe_layer_idx]
def _node_count_with_rank_mapping(
pg: ProcessGroup | StatelessProcessGroup,