如何加载checkpoint继续训练？

合为 pytorch_model.bin后直接加载进来执行fit，会报错flash_att相关的错
        ckpt_path = os.path.join(config['checkpoint_dir'], 'pytorch_model.bin')
        ckpt = torch.load(ckpt_path, map_location='cpu')
        msg = trainer.model.load_state_dict(ckpt, False)
        best_valid_score, best_valid_result = trainer.fit(
            train_loader, None, saved=saved, show_progress=config['show_progress']
        )