wehub-resource-sync
37 lines
809 B
C++
37 lines
809 B
C++
//
|
|
// quantization.hpp
|
|
// MNNConverter
|
|
//
|
|
// Created by MNN on 2020/07/28.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#ifndef MNN_CONVERTER_COMPRESSION_QUANTIZATION_HPP_
|
|
#define MNN_CONVERTER_COMPRESSION_QUANTIZATION_HPP_
|
|
|
|
#include <vector>
|
|
#include <string>
|
|
#include <unordered_map>
|
|
|
|
#include "MNN_compression.pb.h"
|
|
|
|
namespace compression {
|
|
|
|
struct Quantization {
|
|
MNN::Compression::QuantizeParams::RoundMode round_mode;
|
|
|
|
struct TensorParams {
|
|
int32_t nbit;
|
|
std::vector<float> scale;
|
|
float zero_point;
|
|
float clamp_min;
|
|
float clamp_max;
|
|
MNN::Compression::LayerQuantizeParams_QuantMethod method;
|
|
};
|
|
std::unordered_map<std::string, std::vector<TensorParams>> tensors;
|
|
};
|
|
|
|
};
|
|
|
|
#endif // MNN_CONVERTER_COMPRESSION_QUANTIZATION_HPP_
|