LLM Learning Hub

workspace/llm-course/home

Norms

A norm measures the length or magnitude of a vector. The L2 norm (Euclidean) is most common in ML; L1 norm sums absolute values. Norms are used for regularization and normalization.

What is Norms?

A norm measures the length or magnitude of a vector. The L2 norm (Euclidean) is most common in ML; L1 norm sums absolute values. Norms are used for regularization and normalization.

A norm measures the length or magnitude of a vector. The L2 norm (Euclidean) is most common in ML; L1 norm sums absolute values. Norms are used for regularization and normalization.

Where is it used?

LayerNorm in every transformer block normalizes activations using L2 norm; weight decay applies an L2 penalty to gradients during LLM training.

How to build it

Use `torch.norm(v, p=2)` for L2 and `torch.norm(v, p=1)` for L1, and implement LayerNorm manually by subtracting the mean and dividing by the L2 norm of a random vector.

Code

A practical example:

example.pypython