add fix routing for performance test

Signed-off-by: zhky <[email protected]>

Author: 洪炜杰

tests/multicard/test_offline_inference_distributed.py

@@ -61,3 +61,21 @@ def test_models_distributed_DeepSeek():
             distributed_executor_backend="mp",
     ) as vllm_model:
         vllm_model.generate_greedy(example_prompts, max_tokens)
+
+
+def test_models_distributed_ep_DeepSeek():
+    example_prompts = [
+        "vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.",
+        "Briefly describe the major milestones in the development of artificial intelligence from 1950 to 2020.",
+        "Compare and contrast artificial intelligence with human intelligence in terms of processing information.",
+    ]
+    dtype = "half"
+    max_tokens = 5
+    with VllmRunner(
+            "deepseek-ai/DeepSeek-V2-Lite",
+            dtype=dtype,
+            tensor_parallel_size=8,
+            enable_expert_parallel=True,
+            distributed_executor_backend="mp",
+    ) as vllm_model:
+        vllm_model.generate_greedy(example_prompts, max_tokens)

vllm_ascend/ops/fused_moe.py

@@ -88,15 +88,14 @@ def fused_experts_with_mc2(
         0:5]
 
     w1 = w1.transpose(1, 2)
-    expert_token_nums = torch.cumsum(expert_token_nums,
-                                     dim=0,
-                                     dtype=torch.int64)
+
     group_list = expert_token_nums.to(torch.int64)
     gate_up_out_list = torch_npu.npu_grouped_matmul(
         x=[expand_x],
         weight=[w1],
         split_item=2,
-        group_list_type=0,
+        # 1 means count mode, to avoid cumulative operation of the group list
+        group_list_type=1,
         group_type=0,
         group_list=group_list,
     )
@@ -110,7 +109,7 @@ def fused_experts_with_mc2(
         x=[gate_up_out],
         weight=[w2],
         split_item=2,
-        group_list_type=0,
+        group_list_type=1,
         group_type=0,
         group_list=group_list,
     )