SpinQuantQuantizer<T, TInput, TOutput>

Models & Types · AiDotNet.Deployment.Optimization.Quantization.Strategies

SpinQuant quantizer - uses learned rotation matrices to reduce outliers before quantization. Applies orthogonal transformations to weight matrices to minimize quantization error.

Implements: IQuantizer<T, TInput, TOutput>

For Beginners

SpinQuant “rotates” the data in a mathematical sense to spread out outliers more evenly. This makes quantization more accurate because extreme values cause the most problems during compression.

How It Works

How It Works:

Key Features:

Reference: Liu et al., “SpinQuant: LLM quantization with learned rotations” (ICLR 2025)

Constructors

Constructor Summary
SpinQuantQuantizer(QuantizationConfiguration config, int numIterations, double learningRate, int blockSize) Initializes a new instance of the SpinQuantQuantizer.

Properties

Property Summary
IReadOnlyList<string> CalibrationWarnings { get; } Gets any warnings generated during calibration.
QuantizationMode Mode { get; }
bool IsCalibrated { get; set; } Gets whether the quantizer has been calibrated.
int BitWidth { get; }

Methods

Method Summary
IFullModel<T, TInput, TOutput> Quantize(IFullModel<T, TInput, TOutput> model, QuantizationConfiguration config)
double GetScaleFactor(string layerName)
int GetZeroPoint(string layerName)
void Calibrate(IFullModel<T, TInput, TOutput> model, IEnumerable<TInput> calibrationData)