how to Normalize word and Image input array for multi-modal models without loosing any of much information?

Question

Embedded matrix of word encoder has embedded_matrix.max() = 3.5220839977264404 and embedded_matrix.min() = -3.671663999557495 as max and min value.