diff --git a/tests/ut/worker/patch_common/test_patch_sampler.py b/tests/ut/worker/patch_common/test_patch_sampler.py
new file mode 100644
index 0000000..a062a97
--- /dev/null
+++ b/tests/ut/worker/patch_common/test_patch_sampler.py
@@ -0,0 +1,28 @@
+import importlib
+import os
+import unittest
+from unittest import mock
+
+import torch
+from vllm.v1.sample.ops import topk_topp_sampler
+
+
+class TestTopKTopPSamplerOptimize(unittest.TestCase):
+
+    @mock.patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_TOPK_OPTIMIZE": "1"})
+    @mock.patch("torch_npu.npu_top_k_top_p")
+    def test_npu_topk_topp_called_when_optimized(self, mock_npu_op):
+        import vllm_ascend.patch.worker.patch_common.patch_sampler
+        importlib.reload(vllm_ascend.patch.worker.patch_common.patch_sampler)
+
+        mock_npu_op.return_value = (torch.randn(1, 3))
+        sampler = topk_topp_sampler.TopKTopPSampler()
+
+        logits = torch.tensor([[1.0, 2.0, 3.0]])
+        k = torch.tensor([2])
+        p = torch.tensor([0.9])
+        generators = {0: torch.Generator()}
+        generators[0].manual_seed(42)
+
+        sampler.forward_native(logits, generators, k, p)
+        mock_npu_op.assert_called_once_with(logits, p, k)
diff --git a/vllm_ascend/patch/worker/patch_common/patch_sampler.py b/vllm_ascend/patch/worker/patch_common/patch_sampler.py
index a6fbfbc..69fcd69 100644
--- a/vllm_ascend/patch/worker/patch_common/patch_sampler.py
+++ b/vllm_ascend/patch/worker/patch_common/patch_sampler.py
@@ -19,6 +19,7 @@
 from typing import Optional
 
 import torch
+import torch_npu
 from vllm.v1.sample.ops.topk_topp_sampler import TopKTopPSampler, random_sample
 from vllm.v1.sample.sampler import Sampler
 
@@ -48,9 +49,13 @@ def apply_min_p(
 
 def _apply_top_k_top_p(
     logits: torch.Tensor,
-    p: torch.Tensor,
     k: torch.Tensor,
+    p: torch.Tensor,
 ) -> torch.Tensor:
+    if p is not None and k is not None:
+        # npu_top_k_top_p's parameter order is (logits, p, k), not (logits, k, p)
+        return torch_npu.npu_top_k_top_p(logits, p, k)
+
     probs = logits.softmax(dim=-1)
     probs_sort, _ = probs.sort(dim=-1, descending=False)