[Perf] add patch to optimize apply_topk_topp (#1732)

### What this PR does / why we need it?
Performance optimization for apply_top_k_top_p
### Does this PR introduce _any_ user-facing change?
Use VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION to enable this feature
### How was this patch tested?
e2e & ut

















- vLLM version: v0.9.2
- vLLM main:
6a9e6b2abf

Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>

tests/e2e/singlecard/test_offline_inference.py

@@ -21,9 +21,12 @@
 Run `pytest tests/test_offline_inference.py`.
 """
 import os
+from unittest.mock import patch
 
 import pytest
+import vllm  # noqa: F401
 from modelscope import snapshot_download  # type: ignore[import-untyped]
+from vllm import SamplingParams
 from vllm.assets.image import ImageAsset
 
 import vllm_ascend  # noqa: F401
@@ -103,3 +106,24 @@ def test_multimodal(model, prompt_template, vllm_runner):
         vllm_model.generate_greedy(prompts=prompts,
                                    images=images,
                                    max_tokens=64)
+
+
+@patch.dict(os.environ, {"VLLM_ASCEND_ENABLE_TOPK_TOPP_OPTIMIZATION": "1"})
+def test_models_topk() -> None:
+    example_prompts = [
+        "Hello, my name is",
+        "The president of the United States is",
+        "The capital of France is",
+        "The future of AI is",
+    ]
+    sampling_params = SamplingParams(max_tokens=5,
+                                     temperature=0.0,
+                                     top_k=50,
+                                     top_p=0.9)
+
+    with VllmRunner("Qwen/Qwen2.5-0.5B-Instruct",
+                    max_model_len=8192,
+                    dtype="float16",
+                    enforce_eager=True,
+                    gpu_memory_utilization=0.7) as vllm_model:
+        vllm_model.generate(example_prompts, sampling_params)