98.98. GELU Activation

Implement the **GELU (Gaussian Error Linear Unit)** activation function.

GELU Approximation:

gelu(x) = 0.5 * x * (1 + tanh(sqrt(2/pi) * (x + 0.044715 * x^3)))

Example:

gelu(0) → 0.0

gelu(1) → 0.8412

gelu(-1) → -0.1588

**Explanation:** GELU is the activation function used in GPT, BERT, and most modern transformers. It smoothly gates the input: positive values pass through nearly unchanged, while negative values are gradually suppressed (not hard-zeroed like ReLU). This stochastic regularization effect helps in training.

Constraints: