ML//Transformer//layer normalization

Normalize activations across the feature dimension (not the batch)

Normalize activations across the feature dimension (not the batch)

Stabilizes training: prevents values from exploding or vanishing across layers.

Pre-norm (normalize before attention/FFN) vs post-norm (after): pre-norm is more stable for deep models.