ITransformerNetwork<T, TInput, TOutput>

Interfaces · AiDotNet.Interpretability.Interfaces

Interface for transformer networks that support attention visualization.

For Beginners

Transformers use “attention” to focus on different parts of the input when making predictions. This interface provides methods to extract these attention patterns for visualization and analysis.

How It Works

For example, in a text classifier, attention might show which words the model focused on most when making its classification decision.

Methods

Method Summary
List<Tensor<T>> GetAttentionWeights(Tensor<T> input) Gets attention weights from all transformer layers.