[CI] Remove VLLM_TEST_CLEAN_GPU_MEMORY to avoid environment variable pollution that causes unnecessary GPU detection, thereby slowing down test case execution. (#3446)

Signed-off-by: wangyu <410167048@qq.com>
Signed-off-by: [Your Name] <your.email@example.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>

tests/entrypoints/openai_api/test_openai_audio_list_endpoints.py

@@ -9,7 +9,6 @@ Migrated from ``tests/e2e/online_serving/test_voxtral_tts.py`` to colocate with
 import os
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import httpx
 import pytest

tests/entrypoints/openai_api/test_qwen3_tts_speaker_embedding.py

@@ -10,7 +10,6 @@ via pre-computed ECAPA-TDNN embeddings, bypassing ref_audio extraction.
 import os
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import struct
 

tests/entrypoints/openai_api/test_qwen3_tts_websocket.py

@@ -16,7 +16,6 @@ from tests.helpers.runtime import OmniServerParams
 from tests.helpers.stage_config import get_deploy_config_path
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice"
 

tests/entrypoints/openai_api/test_serving_speech_batch.py

@@ -9,7 +9,6 @@ Validates bulk synthesis via the batch API with actual model inference.
 import os
 
 os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
-os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 import base64
 import struct

tests/entrypoints/test_omni_sleep_mode.py

@@ -18,7 +18,10 @@ from vllm_omni.platforms import current_omni_platform
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger("OmniTest")
-pytestmark = [pytest.mark.advanced_model]
+pytestmark = [
+    pytest.mark.advanced_model,
+    pytest.mark.usefixtures("clean_gpu_memory_between_tests"),
+]
 
 
 def clean_gpu_envs():

tests/entrypoints/test_stage_utils.py

@@ -55,7 +55,6 @@ def _make_mock_platform(mocker, device_type: str = "cuda", env_var: str = "CUDA_
 
 @pytest.mark.core_model
 @pytest.mark.cpu
-@pytest.mark.usefixtures("clean_gpu_memory_between_tests")
 def test_set_stage_devices_respects_logical_ids(mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch):
     # Preserve an existing logical mapping and ensure devices "0,1" map through it.
     monkeypatch.setenv("CUDA_VISIBLE_DEVICES", "6,7")
@@ -77,7 +76,6 @@ def test_set_stage_devices_respects_logical_ids(mocker: MockerFixture, monkeypat
 
 @pytest.mark.core_model
 @pytest.mark.cpu
-@pytest.mark.usefixtures("clean_gpu_memory_between_tests")
 def test_set_stage_devices_handles_not_enough_devices(mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch):
     # Preserve an existing logical mapping and ensure devices "0,1" map through it.
     monkeypatch.setenv("CUDA_VISIBLE_DEVICES", "6,7")
@@ -98,7 +96,6 @@ def test_set_stage_devices_handles_not_enough_devices(mocker: MockerFixture, mon
     assert os.environ["CUDA_VISIBLE_DEVICES"] == "6,7"
 
 
-@pytest.mark.usefixtures("clean_gpu_memory_between_tests")
 def test_set_stage_devices_npu_platform(mocker: MockerFixture, monkeypatch: pytest.MonkeyPatch):
     """Test that set_stage_devices works correctly for NPU platform."""
     monkeypatch.setenv("ASCEND_RT_VISIBLE_DEVICES", "4,5")