C# How to pass multiple losses and labels? TensorFlow.NET (TensorFlow.Keras)

Question

I am using C# with TensorFlow.NET (TensorFlow.Keras).

I want to implement alphazero and I want to create a neural network with multiple outputs.

I could create a model with multiple outputs like:

    var model = keras.Model((input), (output1, output2));

However, I do not know how to pass multiple loss function in compile phase and multiple labels in fit phase.

    model.compile(
        optimizer: keras.optimizers.SGD(1e-3f),
        metrics: new string[] { "mse", "accuracy" },
        loss: ??????????
    );


    model.fit(x_train, ?????????????);

In C#, it is not allowed to pass an array as loss nor label, but in Python, it seems that we can use dictionary with curly brackets like:

    model.compile(optimizer='sgd',
        loss={'output1': 'mean_squared_error', 'output2': 'softmax_cross_entropy_with_logits'},
        metrics=['mse','accuracy'])


    labels= {'output1': np.array([row['value'] for row in minibatch]),
                        'output2': np.array([row['AV'] for row in minibatch])} 
    fit = self.model.fit(x_train, lables)

How can we attain this in C#? Thank you.