Artificial Intelligence

Attention Score

The numerical value representing how much one token should focus on another token in the attention mechanism. Higher scores mean stronger relationships between tokens.

Why It Matters

Attention scores are what make transformers interpretable — you can visualize them to see which parts of the input the model is focusing on for each output.

Example

In translating 'black cat,' the word 'noir' (French for black) would have a high attention score with 'black' and a low score with 'cat'.

Think of it like...

Like a heat map showing where a student's eyes focus while reading — bright spots indicate the words they are paying most attention to.

Related Terms