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, 19:45 UTC
[CI/Build] Use a fixed seed to avoid flaky tests (#14480)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Mar 8, 2025 at 19:30 UTC
33f227e16bfc05910c6650a9d22911031d7f67d6
4 files changed
+7
-6
tests/entrypoints/openai/test_chat_echo.py
-2
@@ -24,8 +24,6 @@ def server():
24
"4080",
25
"--chat-template",
26
DUMMY_CHAT_TEMPLATE,
27
- "--seed",
28
- "0",
27
]
28
29
with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
tests/entrypoints/openai/test_metrics.py
-2
@@ -47,8 +47,6 @@ def default_server_args():
47
"--enforce-eager",
48
"--max-num-seqs",
49
"128",
50
- "--seed",
51
- "0",
50
]
51
52
tests/entrypoints/openai/test_root_path.py
-2
@@ -30,8 +30,6 @@ def server():
30
"/" + ROOT_PATH,
31
"--chat-template",
32
DUMMY_CHAT_TEMPLATE,
33
- "--seed",
34
- "0",
33
]
34
envs = os.environ.copy()
35
tests/utils.py
+7
@@ -76,6 +76,7 @@ class RemoteOpenAIServer:
76
vllm_serve_args: list[str],
77
*,
78
env_dict: Optional[dict[str, str]] = None,
79
+ seed: Optional[int] = 0,
80
auto_port: bool = True,
81
max_wait_seconds: Optional[float] = None) -> None:
82
if auto_port:
@@ -87,6 +88,12 @@ class RemoteOpenAIServer:
88
vllm_serve_args = vllm_serve_args + [
89
"--port", str(get_open_port())
90
]
91
+ if seed is not None:
92
+ if "--seed" in vllm_serve_args:
93
+ raise ValueError("You have manually specified the seed "
94
+ f"when `seed={seed}`.")
95
+
96
+ vllm_serve_args = vllm_serve_args + ["--seed", str(seed)]
97
98
parser = FlexibleArgumentParser(
99
description="vLLM's remote OpenAI server.")