Self-Attention

Softmax Normalization

Activation function transforming attention scores into probability distribution, ensuring that the sum of attention weights equals 1 for each position.

← Tillbaka