Keras 我应该在Vision Transformer中进行哪些更改以执行图像重建的自动编码器转换器？

我的代码：

import tensorflow as tf

class Transformer(tf.keras.Model):
    def __init__(self, dim, depth, heads, mlp_dim):
        super().__init__()
        layers = []
        for _ in range(depth):
            layers.extend([
                Residual(PreNorm(dim, Attention(dim, heads = heads))),
                Residual(PreNorm(dim, FeedForward(dim, mlp_dim)))
            ])
        self.net = tf.keras.Sequential(layers)

    def call(self, x):
        return self.net(x)