CalibrationHelper<T, TInput, TOutput>

Helpers & Utilities · AiDotNet.Deployment.Optimization.Quantization.Calibration

Helper class for calibrating quantizers using real forward passes.

For Beginners

Calibration is the process of running sample data through a model to understand the typical range of values (activations) that flow through each layer. This information is crucial for accurate quantization.

How It Works

Why Real Forward Passes Matter:

Supported Models:

Constructors

Constructor Summary
CalibrationHelper(QuantizationConfiguration config) Initializes a new instance of CalibrationHelper.

Methods

Method Summary
ActivationStatistics<T> CollectActivationStatistics(IFullModel<T, TInput, TOutput> model, IEnumerable<TInput> calibrationData) Collects activation statistics by running calibration data through the model.