inference/launch_gradio_demo.py

import sys
sys.path.append('.')

import argparse
import os
import subprocess
from threading import Thread

from inference.interface import PenguinVLQwen3GradioInterface
from inference.server import PenguinVLQwen3PlainClient


def launch_gradio_demo(model_path, server_port=16667, interface_port=33666, server_name="0.0.0.0", nproc=1, example_dir="./assets/inputs"):
    server_thread = Thread(
        target=lambda: subprocess.run(
            [
                sys.executable, "-m",
                "inference.server.plain_server",
                "--model-path", model_path,
                "--nproc", str(nproc),
                "--port", str(server_port),
            ]
        )
    )
    server_thread.daemon = True
    server_thread.start()

    if example_dir is not None and not os.path.isdir(example_dir):
        example_dir = None

    model_client = PenguinVLQwen3PlainClient(port=server_port)
    interface = PenguinVLQwen3GradioInterface(
        model_client,
        example_dir=example_dir,
        server_name=server_name,
        server_port=interface_port,
    )
    interface.launch()


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--model-path", "--model_path", type=str, required=True)
    parser.add_argument("--server-port", "--server_port", type=int, default=16667)
    parser.add_argument("--interface-port", "--interface_port", type=int, default=33666)
    parser.add_argument("--server-name", "--server_name", type=str, default="0.0.0.0")
    parser.add_argument("--nproc", type=int, default=1)
    parser.add_argument("--example-dir", "--example_dir", type=str, default="./assets/inputs")
    args = parser.parse_args()

    launch_gradio_demo(
        model_path=args.model_path,
        server_port=args.server_port,
        interface_port=args.interface_port,
        server_name=args.server_name,
        nproc=args.nproc,
        example_dir=args.example_dir,
    )


if __name__ == "__main__":
    main()