将
grouped_dataset = dataset.group_by_window(
        key_func=lambda seq: smart_length(tf_rank1_tensor_len(seq), bucket_bounds=bounds), # choose a bucket
        reduce_func=lambda key, ds: pad_batch(ds, batch_size, padding=padding, padded_shapes=pad_shape), # apply reduce funtion to pad
        window_size=window_size)
改为：

grouped_dataset = dataset.apply(tf.compat.v1.data.experimental.group_by_window(
        key_func=lambda seq: smart_length(tf_rank1_tensor_len(seq), bucket_bounds=bounds), # choose a bucket
        reduce_func=lambda key, ds: pad_batch(ds, batch_size, padding=padding, padded_shapes=pad_shape), # apply reduce funtion to pad
        window_size=window_size))
即可