diff --git a/vllm_mindspore/model_executor/layers/fused_moe/fused_moe.py b/vllm_mindspore/model_executor/layers/fused_moe/fused_moe.py
index 695c481f38bdd68c1e9feade3e40cacfa35ed37e..764de873882cb1eee8890a945c62d018e7c89cf0 100644
--- a/vllm_mindspore/model_executor/layers/fused_moe/fused_moe.py
+++ b/vllm_mindspore/model_executor/layers/fused_moe/fused_moe.py
@@ -24,6 +24,7 @@ from typing import Optional
 import mindspore as ms
 import numpy as np
 from mindspore import Tensor, mint, nn, ops
+import ms_custom_ops
 from mindspore.ops.auto_generate import (GroupedMatmulV4, MoeDistributeCombine,
                                          MoeDistributeDispatch,
                                          MoeGatingTopKSoftmax,
@@ -92,8 +93,8 @@ class FusedExperts(nn.Cell):
         if moe_config.moe_parallel_config.ep_size > 1 and \
            moe_config.moe_parallel_config.tp_size == 1:
             self.moe_mode = MoeMode.PURE_EP
-            self.dispatch = MoeDistributeDispatch()
-            self.combine = MoeDistributeCombine()
+            self.dispatch = ms_custom_ops.moe_distribute_dispatch_v3
+            self.combine = ms_custom_ops.moe_distribute_combine_v3
             self.dispatch_tp_world_size = 0 if is_910b() else 1
             self.dispatch_shared_expert_num = 0 if is_910b() else 1
             self.max_bs = 256 if is_910b() else 512
@@ -245,4 +246,36 @@ class FusedExperts(nn.Cell):
 
     def run_ep(self, hidden_states, w1, w2, topk_ids, topk_weights, activation,
                global_num_experts):
-        raise NotImplementedError("ep mode not implemented yet.")
+        topk_ids = topk_ids.astype(ms.int32)
+        topk_weights = topk_weights.astype(ms.float32)
+        # Dispatch
+        sorted_input_tensor, _, assist_info_for_combine, group_list, ep_recv_counts, tp_recv_counts, _ = self.dispatch(
+            x=hidden_states,
+            expert_ids=topk_ids,
+            ep_world_size=self.ep_size,
+            ep_rank_id=self.ep_rank,
+            moe_expert_num=global_num_experts,
+            group_ep=self.ep_group,
+            tp_world_size=self.dispatch_tp_world_size,
+            global_bs=self.max_bs)
+
+        # GroupMatmul
+        expert_output = self._ffn(sorted_input_tensor, w1, w2, group_list,
+                                  activation)
+
+        # Combine
+        moe_output = self.combine(
+            expand_x=expert_output,
+            expert_ids=topk_ids,
+            assist_info_for_combine=assist_info_for_combine,
+            ep_send_counts=ep_recv_counts,
+            expert_scales=topk_weights,
+            ep_world_size=self.ep_size,
+            ep_rank_id=self.ep_rank,
+            moe_expert_num=global_num_experts,
+            tp_send_counts=tp_recv_counts,
+            group_ep=self.ep_group,
+            tp_world_size=self.dispatch_tp_world_size,
+            shared_expert_num=self.dispatch_shared_expert_num,
+            global_bs=self.max_bs)
+        return moe_output