Zubnet AILearnWiki › Attention Visualization
Fundamentals

Attention Visualization

Attention Maps, Attention Heatmap
Visualizing what a Transformer model "attends to" by displaying the attention weights as heatmaps. For each query token, the attention map shows how much weight it assigns to every other token. High weights (bright spots) indicate strong attention — the model considers those tokens highly relevant to the current computation.

Why it matters

Attention visualization is the most intuitive way to peek inside a Transformer and understand its reasoning. When a model translates "le chat noir" to "the black cat," attention maps show that "black" attends strongly to "noir" and "cat" to "chat." This helps debug model behavior, understand failures, and build intuition about how attention works.

Deep Dive

The attention weight matrix is (seq_len × seq_len) for each head and layer. To visualize: pick a layer and head, display the matrix as a heatmap where row i shows which tokens token i attends to. Bright cells mean high attention. For multi-head attention, you can visualize individual heads (each specializes in different patterns) or average across heads (overall attention distribution).

What Attention Maps Show (and Don't)

Attention maps show which tokens a head considers when computing its output, but they don't directly show what the model "understands" or why it made a decision. High attention doesn't mean "important" — some heads attend to punctuation or positional patterns without semantic meaning. Attention maps are descriptive (what the model looked at) not explanatory (why it made its decision). They're a useful debugging tool, not a complete explanation.

Tools

BertViz provides interactive attention visualizations for Transformer models. Ecco and Captum offer attention-based interpretability for PyTorch models. For LLMs accessed via API, some providers return attention weights or log-probabilities that enable partial visualization. In image generation, cross-attention maps show which image regions correspond to which prompt words — useful for understanding why the model placed objects where it did.

Related Concepts

← All Terms
ESC