Text this: Enhancing Interpretability of Neural Compact Models: Toward Reliable Device Modeling