harmony 鸿蒙OH_NN_QuantParam
OH_NN_QuantParam
Overview
Quantization information.
In quantization scenarios, the 32-bit floating-point data type is quantized into the fixed-point data type according to the following formula:
s and z are quantization parameters, which are stored by scale and zeroPoint in OH_NN_QuantParam. r is a floating point number, q is the quantization result, q_min is the lower bound of the quantization result, and q_max is an upper bound of a quantization result. The calculation method is as follows:
The clamp function is defined as follows:
Since: 9
Related Modules:
Summary
Member Variables
| Name | Description |
|---|---|
| quantCount | Specifies the length of the numBits, scale, and zeroPoint arrays. In the per-layer quantization scenario, quantCount is usually set to 1. That is, all channels of a tensor share a set of quantization parameters. In the per-channel quantization scenario, quantCount is usually the same as the number of tensor channels, and each channel uses its own quantization parameters. |
| numBits | Number of quantization bits |
| scale | Pointer to the scale data in the quantization formula |
| zeroPoint | Pointer to the zero point data in the quantization formula |
Member Variable Description
numBits
const uint32_t* OH_NN_QuantParam::numBits
Description
Number of quantization bits
quantCount
uint32_t OH_NN_QuantParam::quantCount
Description
Specifies the length of the numBits, scale, and zeroPoint arrays. In the per-layer quantization scenario, quantCount is usually set to 1. That is, all channels of a tensor share a set of quantization parameters. In the per-channel quantization scenario, quantCount is usually the same as the number of tensor channels, and each channel uses its own quantization parameters.
scale
const double* OH_NN_QuantParam::scale
Description
Pointer to the scale data in the quantization formula
zeroPoint
const int32_t* OH_NN_QuantParam::zeroPoint
Description
Pointer to the zero point data in the quantization formula
你可能感兴趣的鸿蒙文章
- 所属分类： 后端技术
- 本文标签：
热门推荐
-
2、 - 优质文章
-
3、 gate.io
-
8、 golang
-
9、 openharmony
-
10、 Vue中input框自动聚焦