A teammate hands you a trained network together with a one-line model card:
Three-layer fully connected network. 12 input features, 4 outputs. Every hidden layer has the same width. 764 learned parameters.
The card counts layers the usual way. The input layer is just the data arriving: it has no neurons, no weights and no biases, and it is not counted as one of the three. Every neuron connects to every neuron in the layer before it with its own weight, and every neuron has its own bias.
How many neurons are in each hidden layer?
Give the exact integer.