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 10, 14:00 UTC
UX: add vLLM env info in '/server_info' (#31899)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Jan 8, 2026 at 01:13 UTC
30399cc725306deff10169ae395864e70b8ab2ba
1 file changed
+19
-3
vllm/entrypoints/serve/instrumentator/server_info.py
+19
-3
@@ -19,6 +19,19 @@ router = APIRouter()
19
PydanticVllmConfig = pydantic.TypeAdapter(VllmConfig)
20
21
22
+def _get_vllm_env_vars():
23
+ from vllm.config.utils import normalize_value
24
+
25
+ vllm_envs = {}
26
+ for key in dir(envs):
27
+ if key.startswith("VLLM_") and "KEY" not in key:
28
+ value = getattr(envs, key, None)
29
+ if value is not None:
30
+ value = normalize_value(value)
31
+ vllm_envs[key] = value
32
+ return vllm_envs
33
+
34
+
35
@router.get("/server_info")
36
async def show_server_info(
37
raw_request: Request,
@@ -26,10 +39,13 @@ async def show_server_info(
39
):
40
vllm_config: VllmConfig = raw_request.app.state.vllm_config
41
server_info = {
29
- "vllm_config": str(vllm_config)
30
- if config_format == "text"
31
- else PydanticVllmConfig.dump_python(vllm_config, mode="json", fallback=str)
42
+ "vllm_config": (
43
+ str(vllm_config)
44
+ if config_format == "text"
45
+ else PydanticVllmConfig.dump_python(vllm_config, mode="json", fallback=str)
46
+ ),
47
# fallback=str is needed to handle e.g. torch.dtype
48
+ "vllm_env": _get_vllm_env_vars(),
49
}
50
return JSONResponse(content=server_info)
51