onert::ir::Quantization Struct Reference

Struct to hold quantization information of a tensor. More...

#include <TypeInfo.h>

Data Fields
std::vector< float >	scales = {0.0f}
std::vector< int32_t >	zero_points = {0}

Detailed Description

Struct to hold quantization information of a tensor.

Note

This struct is used only for quantized tensors.

   For non-quantized tensors, scales, and zero_points have only one element,
   and scale value is zero.

   For per-channel quantized tensors, scales vectors should have same size as number of
   channels of the tensor. And zero_points vector should be same size as scales vector if
   quantization type is assymetric, otherwise it will be ignored.

Definition at line 44 of file TypeInfo.h.

Field Documentation

scales

std::vector<float> onert::ir::Quantization::scales = {0.0f}

Definition at line 46 of file TypeInfo.h.

{0.0f};

Referenced by onert::ir::TypeInfo::quantization(), onert::ir::TypeInfo::quantization(), onert::ir::TypeInfo::scale(), and onert::ir::TypeInfo::scales().

zero_points

std::vector<int32_t> onert::ir::Quantization::zero_points = {0}

Definition at line 47 of file TypeInfo.h.

{0};

Referenced by onert::ir::TypeInfo::quantization(), onert::ir::TypeInfo::quantization(), onert::ir::TypeInfo::zero_point(), and onert::ir::TypeInfo::zero_points().

---

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

• runtime/onert/core/include/ir/TypeInfo.h