Salta para o conte√ļdo

Referência para ultralytics/utils/dist.py

Nota

Este ficheiro est√° dispon√≠vel em https://github.com/ultralytics/ ultralytics/blob/main/ ultralytics/utils/dist .py. Se detectares um problema, por favor ajuda a corrigi-lo contribuindo com um Pull Request ūüõ†ÔłŹ. Obrigado ūüôŹ!



ultralytics.utils.dist.find_free_network_port()

Encontra uma porta livre no localhost.

√Č √ļtil no treino de um √ļnico n√≥ quando n√£o queremos ligar-nos a um n√≥ principal real mas temos de definir o MASTER_PORT vari√°vel de ambiente.

Código fonte em ultralytics/utils/dist.py
def find_free_network_port() -> int:
    """
    Finds a free port on localhost.

    It is useful in single-node training when we don't want to connect to a real main node but have to set the
    `MASTER_PORT` environment variable.
    """
    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
        s.bind(("127.0.0.1", 0))
        return s.getsockname()[1]  # port



ultralytics.utils.dist.generate_ddp_file(trainer)

Gera um ficheiro DDP e devolve o nome do ficheiro.

Código fonte em ultralytics/utils/dist.py
def generate_ddp_file(trainer):
    """Generates a DDP file and returns its file name."""
    module, name = f"{trainer.__class__.__module__}.{trainer.__class__.__name__}".rsplit(".", 1)

    content = f"""
# Ultralytics Multi-GPU training temp file (should be automatically deleted after use)
overrides = {vars(trainer.args)}

if __name__ == "__main__":
    from {module} import {name}
    from ultralytics.utils import DEFAULT_CFG_DICT

    cfg = DEFAULT_CFG_DICT.copy()
    cfg.update(save_dir='')   # handle the extra key 'save_dir'
    trainer = {name}(cfg=cfg, overrides=overrides)
    trainer.args.model = "{getattr(trainer.hub_session, 'model_url', trainer.args.model)}"
    results = trainer.train()
"""
    (USER_CONFIG_DIR / "DDP").mkdir(exist_ok=True)
    with tempfile.NamedTemporaryFile(
        prefix="_temp_",
        suffix=f"{id(trainer)}.py",
        mode="w+",
        encoding="utf-8",
        dir=USER_CONFIG_DIR / "DDP",
        delete=False,
    ) as file:
        file.write(content)
    return file.name



ultralytics.utils.dist.generate_ddp_command(world_size, trainer)

Gera e devolve o comando para a formação distribuída.

Código fonte em ultralytics/utils/dist.py
def generate_ddp_command(world_size, trainer):
    """Generates and returns command for distributed training."""
    import __main__  # noqa local import to avoid https://github.com/Lightning-AI/lightning/issues/15218

    if not trainer.resume:
        shutil.rmtree(trainer.save_dir)  # remove the save_dir
    file = generate_ddp_file(trainer)
    dist_cmd = "torch.distributed.run" if TORCH_1_9 else "torch.distributed.launch"
    port = find_free_network_port()
    cmd = [sys.executable, "-m", dist_cmd, "--nproc_per_node", f"{world_size}", "--master_port", f"{port}", file]
    return cmd, file



ultralytics.utils.dist.ddp_cleanup(trainer, file)

Elimina o ficheiro tempor√°rio, caso tenha sido criado.

Código fonte em ultralytics/utils/dist.py
def ddp_cleanup(trainer, file):
    """Delete temp file if created."""
    if f"{id(trainer)}.py" in file:  # if temp_file suffix in file
        os.remove(file)





Criado em 2023-11-12, Atualizado em 2024-06-02
Autores: glenn-jocher (5), Burhan-Q (1), Laughing-q (1)