ModelRefs / Feed-Forward Network (FFN) — AI Glossary

Feed-Forward Network (FFN) — AI Glossary

The position-wise fully connected sub-layer in a transformer block, responsible for most of the model's stored knowledge. Also called FFN or MLP layer.

Overview

Each transformer layer contains a two-layer MLP applied identically to each token position. The FFN typically expands the hidden dimension 4× before projecting back. SwiGLU (used in LLaMA, PaLM) and GeGLU variants outperform ReLU FFNs. Mechanistic interpretability research suggests FFN layers act as key-value memory stores.

Reference details

Topicarchitecture
Also known asFFN, MLP layer, position-wise FFN
Last reviewed2026-06-24

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Feed-Forward Network (FFN) — AI Glossary.

Frequently asked questions

What is Feed-Forward Network (FFN)?

The position-wise fully connected sub-layer in a transformer block, responsible for most of the model's stored knowledge.

Is Feed-Forward Network (FFN) the same as FFN?

Yes — FFN, MLP layer, position-wise FFN are common aliases for Feed-Forward Network (FFN).

What concepts are related to Feed-Forward Network (FFN)?

Closely related concepts include transformer, residual connection, layer normalization.