QuantizationReport

classin packageklyn.math.nn.quantization

public class QuantizationReport

QuantizationReport

Summary of a successfully published Q8 checkpoint; no tensor data is retained.

Example

print(report.sourceBytes, report.outputBytes, report.maxAbsoluteError)

Properties

Modifier and Type	Member	Description
public readonly property	maxAbsoluteError `maxAbsoluteError as Double`	Maximum scalar reconstruction error; not a model-quality score.
public readonly property	outputBytes `outputBytes as ULong`	Total output checkpoint file bytes, including scales and headers.
public readonly property	preservedTensors `preservedTensors as ULong`	Number of tensors preserved byte-for-byte.
public readonly property	quantizedTensors `quantizedTensors as ULong`	Number of tensors converted to INT8.
public readonly property	sourceBytes `sourceBytes as ULong`	Total source checkpoint file bytes.

Constructors

Modifier and Type	Member	Description
public	QuantizationReport `QuantizationReport(sourceBytes as ULong, outputBytes as ULong, quantizedTensors as ULong, preservedTensors as ULong, maxAbsoluteError as Double)`	Creates a conversion report without owning weight buffers.