L2 Normalization

Implement L2 Normalization for a 2D tensor. L2 normalization is a technique where each row of the input tensor is normalized by the Euclidean (L2) norm of its elements.

The formula for L2 Normalization is:

$$\text{y} = \frac{x}{\sqrt{\sum x_i^2}}$$

where the sum of squared values is computed across the second dimension (D) for each element in the first dimension (B).

Input:

• Tensor $\text{X}$ of shape $(\text{B}, \text{D})$ (input data)

Output:

• Tensor $\text{Y}$ of shape $(\text{B}, \text{D})$ (normalized data)

Notes:

• For numerical stability, you may need to add a small epsilon $\epsilon = 10^{-10}$ to the denominator to avoid division by zero.
• After normalization, the L2 norm of each row should be approximately 1.0.