A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 02:15 UTC
[CI/Build] Fix Basic Models Test (#20728)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Jul 10, 2025 at 17:57 UTC
7571a4a7e5e87c85fd9f84ead184182cd4ff8a9a
1 file changed
+9
tests/models/test_initialization.py
+9
@@ -12,11 +12,20 @@ from vllm.utils import GiB_bytes
12
from vllm.v1.core.kv_cache_utils import get_kv_cache_config
13
from vllm.v1.engine.core import EngineCore as V1EngineCore
14
15
+from ..utils import create_new_process_for_each_test
16
from .registry import HF_EXAMPLE_MODELS
17
18
19
@pytest.mark.parametrize("model_arch", HF_EXAMPLE_MODELS.get_supported_archs())
20
+@create_new_process_for_each_test()
21
def test_can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch):
22
+ """The reason for using create_new_process_for_each_test is to avoid
23
+ the WARNING:
24
+ "We must use the 'spawn' multiprocessing start method. Overriding
25
+ VLLM_WORKER_MULTIPROC_METHOD to 'spawn'."
26
+ The spawn process causes the _initialize_kv_caches_v1 function below to
27
+ become ineffective.
28
+ """
29
model_info = HF_EXAMPLE_MODELS.get_hf_info(model_arch)
30
model_info.check_available_online(on_fail="skip")
31
model_info.check_transformers_version(on_fail="skip")