[AMD][DSV4] moe: enable shared-expert fusion on the grouped-topk path (megamoe) (#40943)
This commit is contained in:
@@ -2556,7 +2556,7 @@ def select_experts(
|
||||
renormalize=renormalize,
|
||||
num_expert_group=num_expert_group,
|
||||
topk_group=topk_group,
|
||||
num_fused_shared_experts=num_fused_shared_experts,
|
||||
num_fused_shared_experts=num_fused_shared_experts_for_gate,
|
||||
routed_scaling_factor=routed_scaling_factor,
|
||||
apply_routed_scaling_factor_on_output=apply_routed_scaling_factor_on_output,
|
||||
scoring_func=scoring_func,
|
||||
@@ -2570,7 +2570,7 @@ def select_experts(
|
||||
renormalize=renormalize,
|
||||
num_expert_group=num_expert_group,
|
||||
topk_group=topk_group,
|
||||
num_fused_shared_experts=num_fused_shared_experts,
|
||||
num_fused_shared_experts=num_fused_shared_experts_for_gate,
|
||||
routed_scaling_factor=routed_scaling_factor,
|
||||
apply_routed_scaling_factor_on_output=apply_routed_scaling_factor_on_output,
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user