[main] adjust the position of warm_up_atb (#2823)

### What this PR does / why we need it?
Adjust the position of warm_up_atb.

### Does this PR introduce _any_ user-facing change?
not involved

### How was this patch tested?
CI passed with existing test.

- vLLM version: main
- vLLM main:
b23fb78623

Signed-off-by: huangxialu <huangxialu1@huawei.com>

tests/ut/worker/test_worker_v1.py

@@ -1009,7 +1009,10 @@ class TestNPUWorker(TestBase):
 
     @patch("vllm_ascend.worker.worker_v1.NPUPlatform.seed_everything")
     @patch("vllm_ascend.worker.worker_v1.logger")
-    def test_compile_or_warm_up_model_with_eager_mode(self, mock_logger,
+    @patch("torch_npu._npu_matmul_add_fp32")
+    def test_compile_or_warm_up_model_with_eager_mode(self,
+                                                      mock_npu_matmul_add,
+                                                      mock_logger,
                                                       mock_seed_everything):
         """Test compile_or_warm_up_model method - eager mode"""
         from vllm_ascend.worker.worker_v1 import NPUWorker
@@ -1051,10 +1054,14 @@ class TestNPUWorker(TestBase):
             # Verify seed setting
             mock_seed_everything.assert_called_once_with(12345)
 
+            # Verify calls
+            mock_npu_matmul_add.assert_called_once()
+
     @patch("vllm_ascend.worker.worker_v1.NPUPlatform.seed_everything")
     @patch("vllm_ascend.worker.worker_v1.logger")
+    @patch("torch_npu._npu_matmul_add_fp32")
     def test_compile_or_warm_up_model_with_graph_capture(
-            self, mock_logger, mock_seed_everything):
+            self, mock_npu_matmul_add, mock_logger, mock_seed_everything):
         """Test compile_or_warm_up_model method - with graph capture enabled"""
         from vllm_ascend.worker.worker_v1 import NPUWorker
 
@@ -1087,6 +1094,9 @@ class TestNPUWorker(TestBase):
             # Verify seed setting
             mock_seed_everything.assert_called_once_with(67890)
 
+            # Verify calls
+            mock_npu_matmul_add.assert_called_once()
+
     @patch("vllm_ascend.worker.worker_v1.CaMemAllocator")
     def test_initialize_from_config_with_sleep_mode(self,
                                                     mock_allocator_class):