LLMTime<T>
Models & Types · AiDotNet.Finance.Forecasting.Foundation
LLM-Time — Zero-Shot Time Series Forecasting via LLM Tokenization.
Inheritance: NeuralNetworkBase<T> → FinancialModelBase<T> → ForecastingModelBase<T> → TimeSeriesFoundationModelBase<T> → LLMTime<T>
For Beginners
LLM-Time takes an unconventional approach: it converts your numerical data into text (like “3.2, 3.5, 3.8, …”) and asks a large language model to predict the next numbers, just like predicting the next word in a sentence. Remarkably, this works without any training on time series data at all. The LLM already understands number patterns from its language training.
How It Works
LLM-Time converts numeric time series into text strings and uses pretrained LLMs (GPT-3, LLaMA) for zero-shot forecasting by treating the task as next-token prediction on numerical text. No fine-tuning is required—the LLM backbone is frozen.
Reference: Gruver et al., “Large Language Models Are Zero-Shot Time Series Forecasters”, NeurIPS 2023.
Example
// Create an LLMTime model for zero-shot forecasting via LLM tokenization
// Converts numeric time series to text strings and uses a frozen LLM for prediction
var architecture = new NeuralNetworkArchitecture<double>(
inputType: InputType.OneDimensional,
taskType: NeuralNetworkTaskType.Regression,
inputHeight: 512, inputWidth: 1, inputDepth: 1, outputSize: 24);
// Training mode with text-based time series representation
var model = new LLMTime<double>(architecture);
// ONNX inference mode with pre-trained model
var onnxModel = new LLMTime<double>(architecture, "llmtime.onnx");
Constructors
| Constructor | Summary |
|---|---|
LLMTime(NeuralNetworkArchitecture<T> architecture, LLMTimeOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) |
Creates an LLMTime model in native mode for training or fine-tuning. |
LLMTime(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, LLMTimeOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) |
Creates an LLMTime model using a pretrained ONNX model. |
Properties
| Property | Summary |
|---|---|
FoundationModelSize ModelSize { get; } |
|
bool IsChannelIndependent { get; } |
|
bool SupportsTraining { get; } |
|
bool UseNativeMode { get; } |
|
int MaxContextLength { get; } |
|
int MaxPredictionHorizon { get; } |
|
int NumFeatures { get; } |
|
int PatchSize { get; } |
|
int PredictionHorizon { get; } |
|
int SequenceLength { get; } |
|
int Stride { get; } |
Methods
| Method | Summary |
|---|---|
Dictionary<string, T> Evaluate(Tensor<T> predictions, Tensor<T> actuals) |
|
Dictionary<string, T> GetFinancialMetrics() |
|
Dictionary<string, Tensor<T>> GetNamedLayerActivations(Tensor<T> input) |
|
ModelMetadata<T> GetModelMetadata() |
|
Tensor<T> ApplyInstanceNormalization(Tensor<T> input) |
|
Tensor<T> AutoregressiveForecast(Tensor<T> input, int steps) |