it seems that it will never use gpus when use pytorch for training

[faf4r]

if the framework is pytorch, it will from donkeycar.parts.pytorch.torch_train import train and use it to train.
in this train function, it has following codes

    if torch.cuda.is_available():
        print('Using CUDA')
        gpus = -1
    else:
        print('Not using CUDA')
        gpus = 0

however, when creating the trainer, it uses cpu flag and never use gpu

    trainer = pl.Trainer(accelerator='cpu', logger=logger,
                         max_epochs=cfg.MAX_EPOCHS, default_root_dir=output_dir)

so I combined them and can use gpus successfully

    if torch.cuda.is_available():
        print('Using CUDA')
        trainer = pl.Trainer(accelerator='gpu', devices=-1, logger=logger,
                         max_epochs=cfg.MAX_EPOCHS, default_root_dir=output_dir)
    else:
        print('Not using CUDA')
        trainer = pl.Trainer(accelerator='cpu', logger=logger,
                         max_epochs=cfg.MAX_EPOCHS, default_root_dir=output_dir)

If it serves no other purpose, I think it's a bug