[Bugfix] Fix shared expert input for latent MoE in EP+DP (Nemotron-H) (#34087)

Signed-off-by: Tomer Natan <tbarnatan@nvidia.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
TomerBN-Nvidia
2026-02-09 16:44:18 +00:00
committed by GitHub
co-authored by Cursor
parent d4f123cc48
commit 995bbf38f1
6 changed files with 30 additions and 3 deletions
@@ -1023,6 +1023,7 @@ class Fp8MoEMethod(FusedMoEMethodBase):
global_num_experts=layer.global_num_experts,
expert_map=layer.expert_map,
apply_router_weight_on_input=layer.apply_router_weight_on_input,
shared_experts_input=layer._get_shared_experts_input(x),
)