Reorder loop in shared expert weight loading (#5719)

python/sglang/srt/models/deepseek_nextn.py

@@ -215,11 +215,11 @@ class DeepseekV3ForCausalLMNextN(DeepseekV3ForCausalLM):
                     "up_proj.weight_scale_inv",
                 ]
             names_to_remove = []
-            for num_repeat in range(self.n_share_experts_fusion):
             for suffix in suffix_list:
                 shared_expert_weight_name = (
                     f"model.layers.0.mlp.shared_experts.{suffix}"
                 )
+                for num_repeat in range(self.n_share_experts_fusion):
                     weights_list.append(
                         (
                             f"model.layers.0."

python/sglang/srt/models/deepseek_v2.py

@@ -1650,11 +1650,11 @@ class DeepseekV2ForCausalLM(nn.Module):
                 desc=f"Cloning {self.n_share_experts_fusion} "
                 "replicas of the shared expert into MoE",
             ):
-                for num_repeat in range(self.n_share_experts_fusion):
                 for suffix in suffix_list:
                     shared_expert_weight_name = (
                         f"model.layers.{moe_layer}.mlp.shared_experts.{suffix}"
                     )
+                    for num_repeat in range(self.n_share_experts_fusion):
                         weights_list.append(
                             (
                                 f"model.layers.{moe_layer}."