Layer
One floor of the model — a repeatable block of math the input passes through on its way up.
Think of it like
An assembly-line station: each does a small transform and hands the part to the next; stack enough and a raw blank becomes a finished thing.
Example
GPT-style models stack dozens of identical layers; “deeper” usually just means more of them bolted on top.
How it actually works
In a transformer, each layer is attention plus a feed-forward network, wrapped in residual connections and normalization. The input’s representation gets refined a little at every layer — early ones catch surface patterns, later ones catch abstract meaning.
For product teams
More layers roughly means more capacity and more cost; depth is one of the dials behind “bigger model.”
For engineers
A single transformer block: multi-head self-attention + FFN + residual + norm, repeated N times.
Related
- Activation function — Made of neurons using an activation function.
- Deep learning — Stacking many is deep learning.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome