do NOT delete the function keras from the variable model, use the variable as it is and do NOT change it to sequential only

Problem $1(1-1)$: Build an MLP network with $8$ neurons in the first hidden layer, $4$ neurons each in the second and third hidden layers. Use $6$ neurons for the output layer. All the hidden layer neurons use ReLU activation function. The output layer neurons use softmax activation function. The network should be compiled using Adam optimizer, categorical$\_$crossentropy loss function and accuracy as a metric.