biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[MISC][Bugfix] Use less CPU when message queue has been empty for some time (#16226)

Browse Source 
Signed-off-by: Povilas Kanapickas <povilas@radix.lt>
This commit is contained in:
Povilas Kanapickas
2025-06-05 19:53:08 +03:00
committed by GitHub
parent 61059bee40
commit 85e2b7bb13
3 changed files with 70 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
tests/basic_correctness/test_basic_correctness.py
View File

@@ -128,15 +128,21 @@ def test_models(
@multi_gpu_test(num_gpus=2)
@pytest.mark.parametrize(
"model, distributed_executor_backend, attention_backend, "
"test_suite", [
("distilbert/distilgpt2", "ray", "", "L4"),
("distilbert/distilgpt2", "mp", "", "L4"),
("meta-llama/Llama-3.2-1B-Instruct", "ray", "", "L4"),
("meta-llama/Llama-3.2-1B-Instruct", "mp", "", "L4"),
("distilbert/distilgpt2", "ray", "", "A100"),
("distilbert/distilgpt2", "mp", "", "A100"),
("distilbert/distilgpt2", "mp", "FLASHINFER", "A100"),
("meta-llama/Meta-Llama-3-8B", "ray", "FLASHINFER", "A100"),
"test_suite, extra_env", [
("distilbert/distilgpt2", "ray", "", "L4", {}),
("distilbert/distilgpt2", "mp", "", "L4", {}),
("distilbert/distilgpt2", "ray", "", "L4", {
"VLLM_SLEEP_WHEN_IDLE": "1"
}),
("distilbert/distilgpt2", "mp", "", "L4", {
"VLLM_SLEEP_WHEN_IDLE": "1"
}),
("meta-llama/Llama-3.2-1B-Instruct", "ray", "", "L4", {}),
("meta-llama/Llama-3.2-1B-Instruct", "mp", "", "L4", {}),
("distilbert/distilgpt2", "ray", "", "A100", {}),
("distilbert/distilgpt2", "mp", "", "A100", {}),
("distilbert/distilgpt2", "mp", "FLASHINFER", "A100", {}),
("meta-llama/Meta-Llama-3-8B", "ray", "FLASHINFER", "A100", {}),
])
@pytest.mark.parametrize("enable_prompt_embeds", [True, False])
def test_models_distributed(
@@ -148,6 +154,7 @@ def test_models_distributed(
distributed_executor_backend: str,
attention_backend: str,
test_suite: str,
extra_env: dict[str, str],
enable_prompt_embeds: bool,
) -> None:
@@ -173,6 +180,9 @@ def test_models_distributed(
attention_backend,
)
for k, v in extra_env.items():
monkeypatch_context.setenv(k, v)
dtype = "half"
max_tokens = 5