Add QAT support for Concatenate layer

Referencing @nutsiepully in #372

We are ramping up support for layers based on feedback from users. So thank you for that. Most of these layers are quite simple, and haven’t been added due to conversion support.

We’re using the NoOpQuantizeConfig workaround for now, but in the future we would love to see native support for the Keras Concatenate layer. Thanks!

[nutsiepully]

@willbattel - Since support for Concatenate is too broad, please add your use case here which doesn’t work once you get a chance and re-open the bug. I’ll close it for now.

[HOLYlmx]

@nutsiepully Hi, I have also get the error that

to_annotate can only be a tf.keras.layers.Layer instance. You passed an instance of type: Tensor’',

and my code is set as your instructions that:

tf.keras.layers.concatenate([conv2,up1])(conv3) returns a tensor. tf.keras.layers.concatenate([conv2,up1]) returns a layer.

The code is as follows:

class NoOpQuantizeConfig(quantize_config.QuantizeConfig):
    """QuantizeConfig which does not quantize any part of the layer."""
    def get_weights_and_quantizers(self, layer):
        return []
    def get_activations_and_quantizers(self, layer):
        return []
    def set_quantize_weights(self, layer, quantize_weights):
        pass
    def set_quantize_activations(self, layer, quantize_activations):
        pass
    def get_output_quantizers(self, layer):
        return []
    def get_config(self):
        return {}
quantize_config = NoOpQuantizeConfig()

x2 = tf.keras.Input(shape=(28,28))
uu = tf.keras.layers.Reshape((28,28,1))(x2)
uu = tf.keras.layers.Conv2D(filters=12, kernel_size=(3, 3), activation='relu')(uu)
uu = tf.keras.layers.Conv2D(filters=12, kernel_size=(3, 3))(uu)

z=tfmot.quantization.keras.quantize_annotate_layer(to_annotate=tf.keras.layers.concatenate([uu,tf.zeros((0,24,24,12))]),quantize_config=quantize_config)

uu = tf.keras.layers.ReLU()(z)
uu = tfmot.quantization.keras.quantize_annotate_layer(tf.keras.layers.UpSampling2D((2,2),interpolation='bilinear'), quantize_config=quantize_config)(uu)
uu = tf.keras.layers.MaxPooling2D(pool_size=(2, 2))(uu)
uu = tf.keras.layers.MaxPooling2D(pool_size=(2, 2))(uu)
uu = tf.keras.layers.Flatten()(uu)
out = tf.keras.layers.Dense(10)(uu)

model = Model(x2,out)

The code was going well when I didnt put the "tf.keras.layer.concatenate()"layer. Meanwhile, I haven’t fount the “concatenate_conifg” parameter mentioned above. Hoping for your help, thank you so much^^