QuantizeSpec Struct

INT8 (or other low-precision) quantization parameters. More...

Declaration

struct simaai::neat::QuantizeSpec { ... }

Included Headers

#include <PreprocessPlan.h>

Public Member Attributes Index

AutoFlag	enable = AutoFlag::Auto
	On/off/auto. More...
int	zero_point = 0
	Zero-point offset (q = round(x/scale) + zero_point). More...
double	scale = 0.0
	Scale factor; 0 means "use the model's calibration". More...
std::string	output_dtype
	Target dtype name (e.g., "int8", "uint8"). More...

Description

INT8 (or other low-precision) quantization parameters.

Definition at line 141 of file PreprocessPlan.h.

Public Member Attributes

enable

AutoFlag simaai::neat::QuantizeSpec::enable = AutoFlag::Auto

On/off/auto.

Definition at line 142 of file PreprocessPlan.h.

142 AutoFlag enable = AutoFlag::Auto;

output_dtype

std::string simaai::neat::QuantizeSpec::output_dtype

Target dtype name (e.g., "int8", "uint8").

Definition at line 145 of file PreprocessPlan.h.

145 std::string output_dtype;

scale

double simaai::neat::QuantizeSpec::scale = 0.0

Scale factor; 0 means "use the model's calibration".

Definition at line 144 of file PreprocessPlan.h.

144 double scale = 0.0;

zero_point

int simaai::neat::QuantizeSpec::zero_point = 0

Zero-point offset (q = round(x/scale) + zero_point).

Definition at line 143 of file PreprocessPlan.h.

143 int zero_point = 0;

---

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

• PreprocessPlan.h

---

Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.