vllm_omni.engine.cuda_mps ¶

Private NVIDIA MPS lifetime for explicitly opted-in local stage processes.

logger `module-attribute` ¶

logger = init_logger(__name__)

Own a private control socket; never stop an operator's existing daemon.

env = {
    "CUDA_VISIBLE_DEVICES": gpu_uuid,
    "CUDA_MPS_PIPE_DIRECTORY": str(root / "pipe"),
    "CUDA_MPS_LOG_DIRECTORY": str(root / "log"),
}

operator_pipe_directory = (
    os.environ.get("CUDA_MPS_PIPE_DIRECTORY")
    if pipe_directory is None
    else pipe_directory
) or None

close() -> None

physical_gpu_uuid(device: str) -> str

Resolve a physical ordinal/UUID through NVML without initializing CUDA.