From 9e6746b3c7b4ec9bcc234db9a654df8eca5781ce Mon Sep 17 00:00:00 2001 From: Jiangyun Zhu Date: Sat, 25 Jul 2026 04:45:09 +0800 Subject: [PATCH] [CI] Stabilize memory-sensitive compile and structured output tests (#49749) Signed-off-by: zjy0516 Co-authored-by: OpenAI Codex --- tests/compile/fullgraph/test_basic_correctness.py | 2 ++ tests/entrypoints/llm/test_struct_output_generate.py | 2 ++ 2 files changed, 4 insertions(+) diff --git a/tests/compile/fullgraph/test_basic_correctness.py b/tests/compile/fullgraph/test_basic_correctness.py index 35989dcde1..7471564340 100644 --- a/tests/compile/fullgraph/test_basic_correctness.py +++ b/tests/compile/fullgraph/test_basic_correctness.py @@ -64,6 +64,8 @@ class TestSetting: "bfloat16", "--max-model-len", "2048", + "--gpu-memory-utilization", + "0.98", ], pp_size=1, tp_size=1, diff --git a/tests/entrypoints/llm/test_struct_output_generate.py b/tests/entrypoints/llm/test_struct_output_generate.py index 3ece272343..219ee7cd38 100644 --- a/tests/entrypoints/llm/test_struct_output_generate.py +++ b/tests/entrypoints/llm/test_struct_output_generate.py @@ -212,6 +212,7 @@ class CarDescription(BaseModel): PARAMS_MODELS_BACKENDS_TOKENIZER_MODE, ) def test_structured_output( + request: pytest.FixtureRequest, backend: str, tokenizer_mode: str, model_name: str, @@ -242,6 +243,7 @@ def test_structured_output( speculative_config=speculative_config, **platform_args, ) + request.addfinalizer(llm.llm_engine.engine_core.shutdown) # # Test 1: Generate JSON output based on a provided schema