深度学习随记

当模型被封装在 nn.DataParallel 中时需要注意的

当模型被封装在 nn.DataParallel 中时，我们需要使用模型的 module 属性来获得真正的模型状态字典。（model.module.state_dict()）

torch.save(
    {
        "model": model.module.state_dict(),
        "optimizer": optimizer._optimizer.state_dict(),
    },
    os.path.join(
        train_config["path"]["ckpt_path"],
        "{}.pth.tar".format(step),
    ),
)

获取某一训练好的模型的输出的自定义方法

def get_speakermodel(mel):
    model = MultiTaskModel().to(device)
    ckpt_path = "/home/nicola/LA_SE/output/ckpt/300000.pth.tar"
    ckpt = torch.load(ckpt_path)
    model.load_state_dict(ckpt["model"])
    with torch.no_grad():
        s_id = model(mel)
    return s_id

微调

    model = FastSpeech2(preprocess_config, model_config).to(device)
    if train:
        ckpt_path = os.path.join(
            train_config["path"]["ckpt_path"],
            "{}.pth.tar".format(args.restore_step),
        )
        ckpt = torch.load(ckpt_path)
        scheduled_optim = ScheduledOptim(
            model, train_config, model_config, args.restore_step
        )
        if args.restore_step:    #将这一句注释掉或者设置为True就是微调了
            scheduled_optim.load_state_dict(ckpt["optimizer"])
        model.train()
        return model, scheduled_optim