Module calibrate 

Calibration pipeline for quantization.

Provides infrastructure for loading calibration datasets, collecting activation histograms, and computing optimal quantization parameters using various calibration methods (MinMax, Percentile, KL-divergence).

Structs

CalibrationDataset

A set of calibration samples loaded from binary files.

CalibrationResult

Results from running the calibration pipeline on a module.

HistogramCollector

Histogram collector that aggregates activation values across calibration samples.

TensorHistogram

Histogram of activation values for a single tensor.

Enums

CalibrationError

Errors that can occur during calibration.

CalibrationMethod

Calibration method for determining quantization parameters.

Functions

calibrate

Compute quantization parameters for a histogram using the specified method.

calibrate_kl_divergence

Compute quantization parameters using KL-divergence (entropy) calibration.

calibrate_minmax

Compute quantization parameters using MinMax calibration.

calibrate_percentile

Compute quantization parameters using Percentile calibration.

per_channel_quantize

Compute per-channel quantization parameters for a weight tensor.

run_calibration

Run the full calibration pipeline on a calibration dataset.