正在嘗試使用https://github.com/jzuern/cifar-classifier中的代碼在 Google Colab 上實現 ResNet-CIFAR 10 模型。我使用的是我自己的自定義激活函數，而不是 ReLU 激活。這是代碼： def fonlaaf(x):     return x/(1-tf.exp(-x)) def resnet_layer(inputs,                 num_filters=16,                 kernel_size=3,                 strides=1, activation='fonlaaf',                 batch_normalization=True,                 conv_first=True):    """2D Convolution-Batch Normalization-Activation stack builder    # Arguments        inputs (tensor): input tensor from input image or previous layer        num_filters (int): Conv2D number of filters        kernel_size (int): Conv2D square kernel dimensions        strides (int): Conv2D square stride dimensions        activation (string): activation name        batch_normalization (bool): whether to include batch normalization        conv_first (bool): conv-bn-activation (True) or            bn-activation-conv (False)    # Returns        x (tensor): tensor as input to the next layer    """    conv = Conv2D(num_filters,                  kernel_size=kernel_size,                  strides=strides,                  padding='same',                  kernel_initializer='he_normal',                  kernel_regularizer=tf.keras.regularizers.l2(1e-4))    x = inputs    if conv_first:        x = conv(x)        if batch_normalization:            x = BatchNormalization()(x)        if activation is not None:            x = fonlaaf(x)    else:        if batch_normalization:            x = BatchNormalization()(x)        if activation is not None:            x = fonlaaf(x)        x = conv(x)    return x得到以下錯誤：ValueError：模型的輸出張量必須是 TensorFlow 的輸出Layer（因此保存過去的層元數據）。找到：Tensor("dense/Softmax:0", shape=(?, 10), dtype=float32)以前對這個錯誤的回答大多沒有奏效。我在這里想念什么？