signet::forge::Dequantizer Class Reference

Dequantizes INT8/INT4 quantized vectors back to float32. More...

#include <quantized_vector.hpp>

Public Member Functions
	Dequantizer (QuantizationParams params)
	Construct a dequantizer with the given parameters.
void	dequantize (const uint8_t *input, float *output) const
	Dequantize a single quantized vector to float32.
std::vector< std::vector< float > >	dequantize_batch (const uint8_t *input, size_t num_vectors) const
	Dequantize a batch of vectors, returning a vector-of-vectors.
std::vector< float >	dequantize_flat (const uint8_t *input, size_t num_vectors) const
	Flat batch dequantize: returns all floats in one contiguous buffer.
const QuantizationParams &	params () const
	Access the quantization parameters.
uint64_t	anomaly_count () const
	EU AI Act Art.12 anomaly tracking: number of dequantized values that fell outside the representable quantization range and were clamped.

Detailed Description

Dequantizes INT8/INT4 quantized vectors back to float32.

Uses SIMD acceleration (AVX2, SSE2, or NEON) when available, with a scalar fallback.

See also

QuantizationParams, Quantizer

Definition at line 198 of file quantized_vector.hpp.

Constructor & Destructor Documentation

Dequantizer()

signet::forge::Dequantizer::Dequantizer ( QuantizationParams  params )

inlineexplicit

Construct a dequantizer with the given parameters.

Definition at line 201 of file quantized_vector.hpp.

Member Function Documentation

anomaly_count()

uint64_t signet::forge::Dequantizer::anomaly_count ( ) const

inline

EU AI Act Art.12 anomaly tracking: number of dequantized values that fell outside the representable quantization range and were clamped.

Definition at line 231 of file quantized_vector.hpp.

dequantize()

void signet::forge::Dequantizer::dequantize ( const uint8_t *  input, float *  output  ) const

inline

Dequantize a single quantized vector to float32.

Parameters

input	Pointer to bytes_per_vector() quantized bytes.
output	Destination buffer; must hold at least dim floats.

Definition at line 1172 of file quantized_vector.hpp.

dequantize_batch()

std::vector< std::vector< float > > signet::forge::Dequantizer::dequantize_batch ( const uint8_t *  input, size_t  num_vectors  ) const

inline

Dequantize a batch of vectors, returning a vector-of-vectors.

Parameters

input	Pointer to num_vectors * bytes_per_vector() quantized bytes.
num_vectors	Number of vectors to dequantize.

Returns

A vector of float vectors, each of size dim.

Definition at line 1206 of file quantized_vector.hpp.

dequantize_flat()

std::vector< float > signet::forge::Dequantizer::dequantize_flat ( const uint8_t *  input, size_t  num_vectors  ) const

inline

Flat batch dequantize: returns all floats in one contiguous buffer.

Parameters

input	Pointer to num_vectors * bytes_per_vector() quantized bytes.
num_vectors	Number of vectors to dequantize.

Returns

A flat float buffer of size num_vectors * dim.

Definition at line 1222 of file quantized_vector.hpp.

params()

const QuantizationParams & signet::forge::Dequantizer::params ( ) const

inline

Access the quantization parameters.

Definition at line 227 of file quantized_vector.hpp.

---

The documentation for this class was generated from the following file:

• include/signet/ai/quantized_vector.hpp