[CI] Stabilize memory-sensitive compile and structured output tests (#49749)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com> Co-authored-by: OpenAI Codex <noreply@openai.com>
Jiangyun Zhu committed
Jul 25, 2026 at 04:45 UTC
9e6746b3c7b4ec9bcc234db9a654df8eca5781ce
2 files changed
+4
tests/compile/fullgraph/test_basic_correctness.py
+2
@@ -64,6 +64,8 @@ class TestSetting:
64
"bfloat16",
65
"--max-model-len",
66
"2048",
67
+ "--gpu-memory-utilization",
68
+ "0.98",
69
],
70
pp_size=1,
71
tp_size=1,
tests/entrypoints/llm/test_struct_output_generate.py
+2
@@ -212,6 +212,7 @@ class CarDescription(BaseModel):
212
PARAMS_MODELS_BACKENDS_TOKENIZER_MODE,
213
)
214
def test_structured_output(
215
+ request: pytest.FixtureRequest,
216
backend: str,
217
tokenizer_mode: str,
218
model_name: str,
@@ -242,6 +243,7 @@ def test_structured_output(
243
speculative_config=speculative_config,
244
**platform_args,
245
)
246
+ request.addfinalizer(llm.llm_engine.engine_core.shutdown)
247
248
#
249
# Test 1: Generate JSON output based on a provided schema