LLM Server Helper

Use start_vllm_servers() to start one local vLLM server for each URL in config.remote.server_urls. With the default configuration, this starts one server.

matchminer_ai.llm.vllm_server

Helpers for starting OpenAI-compatible local vLLM servers.

start_vllm_servers

start_vllm_servers(*, config: MMAIConfig | None = None, task: str = 'patient', extra_args: Sequence[str] | None = None, stdout: int | None = None, stderr: int | None = None, print_url: bool = True, wait_until_ready: bool = True, ready_timeout: float = 600.0, ready_poll_interval: float = 5.0) -> list[subprocess.Popen[str]]

Start one local vLLM server for each URL in config.remote.server_urls.

Source code in src/matchminer_ai/llm/vllm_server.py

def start_vllm_servers(
    *,
    config: MMAIConfig | None = None,
    task: str = "patient",
    extra_args: Sequence[str] | None = None,
    stdout: int | None = None,
    stderr: int | None = None,
    print_url: bool = True,
    wait_until_ready: bool = True,
    ready_timeout: float = 600.0,
    ready_poll_interval: float = 5.0,
) -> list[subprocess.Popen[str]]:
    """Start one local vLLM server for each URL in ``config.remote.server_urls``."""
    resolved_config = config or load_default_preset()
    return [
        start_vllm_server(
            config=resolved_config,
            task=task,
            server_index=server_index,
            extra_args=extra_args,
            stdout=stdout,
            stderr=stderr,
            print_url=print_url,
            wait_until_ready=wait_until_ready,
            ready_timeout=ready_timeout,
            ready_poll_interval=ready_poll_interval,
        )
        for server_index, _url in enumerate(_remote_server_urls(resolved_config))
    ]