Zubnet AILearnWiki › Self-Attention
Fundamentals

Self-Attention

Scaled Dot-Product Attention
An attention mechanism where a sequence attends to itself — every token computes its relevance to every other token in the same sequence. The queries, keys, and values all come from the same input. This lets each token gather information from all other tokens, weighted by relevance. Self-attention is the core operation in every Transformer layer.

Why it matters

Self-attention is what makes Transformers work. It replaced the sequential processing of RNNs with parallel, direct connections between all positions. The word "bank" in "river bank" attends to "river" to resolve its meaning, regardless of how far apart they are. This ability to directly connect any two positions is why Transformers handle long-range dependencies so well.

Deep Dive

The computation: for input X, compute Q = X·W_Q, K = X·W_K, V = X·W_V. Then: Attention(Q,K,V) = softmax(Q·K^T / √d_k) · V. The softmax(Q·K^T) produces an N×N attention matrix where entry (i,j) represents how much token i attends to token j. The √d_k scaling prevents dot products from growing too large in high dimensions, which would push softmax into saturated regions with near-zero gradients.

Causal vs. Bidirectional

In decoder-only LLMs (GPT, Claude, Llama), self-attention is causal: each token can only attend to previous tokens (including itself). This is enforced by a causal mask that sets future positions to −∞ before softmax. In encoder models (BERT), self-attention is bidirectional: every token attends to every other token. The causal constraint is what makes autoregressive generation possible — the model can't "peek" at future tokens.

The Quadratic Cost

Self-attention computes an N×N attention matrix, making it O(N²) in both time and memory. For a 128K token context, that's ~16 billion entries per layer per head. This quadratic scaling is the fundamental limitation that drives research into sparse attention, linear attention, Flash Attention (which reduces memory but not compute), and SSMs (which avoid the N×N matrix entirely). Every approach to long-context modeling is ultimately about managing this quadratic cost.

Related Concepts

← All Terms
← Scaling Laws Self-Supervised Learning →