GLALanguageModel<T>

Models & Types · AiDotNet.NeuralNetworks

Implements a full GLA (Gated Linear Attention) language model: embedding + N GLA blocks + RMS norm + LM head.

Inheritance: NeuralNetworkBase<T> → DeclaredModelLayoutBase<T> → TokenLanguageModelLayoutBase<T> → GLALanguageModel<T>

For Beginners

GLA is an efficient attention mechanism that uses gates to control information flow, achieving Transformer-level quality with much lower computational cost.

How It Works

GLA introduces hardware-efficient gated linear attention with data-dependent gating and chunk-wise parallel computation. It matches Transformer quality with sub-quadratic complexity.

Reference: Yang et al., “Gated Linear Attention Transformers with Hardware-Efficient Training”, 2024.

Example

var architecture = new NeuralNetworkArchitecture<float>(inputFeatures: 8, outputSize: 4);
var tokens = Tensor<float>.CreateRandom(new[] { 1, 128 });
var trainX = Tensor<float>.CreateRandom(4, 8);
var trainY = Tensor<float>.CreateRandom(4, 2);
var result = new AiModelBuilder<float, Tensor<float>, Tensor<float>>()
    .ConfigureModel(new GLALanguageModel<float>(architecture))
    .Build(trainX, trainY);
var logits = result.Predict(tokens);

Constructors

Constructor Summary
GLALanguageModel(NeuralNetworkArchitecture<T> architecture, GLAOptions options, ILossFunction<T> lossFunction, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)

Properties

Property Summary
bool SupportsTraining { get; }
int ModelDimension { get; } Gets the model dimension (d_model).
int NumLayers { get; } Gets the number of GLA blocks.
int VocabSize { get; } Gets the vocabulary size.

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()