1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
|
#pragma once
#include "caffe2/quantization/server/dnnlowp.h"
namespace dnnlowp {
class QuantizationErrorMinimization {
public:
virtual TensorQuantizationParams ChooseQuantizationParams(
const Histogram& hist,
bool preserve_sparsity = false,
int precision = 8) = 0;
virtual ~QuantizationErrorMinimization(){};
};
class NormMinimization : public QuantizationErrorMinimization {
public:
enum Kind {
L1,
L2,
};
NormMinimization(Kind kind) : kind_(kind) {}
/**
* Faster approximate search
*/
TensorQuantizationParams NonlinearQuantizationParamsSearch(
const Histogram& hist,
bool preserve_sparsity = false,
int precision = 8);
TensorQuantizationParams ChooseQuantizationParams(
const Histogram& hist,
bool preserve_sparsity = false,
int precision = 8) override;
protected:
Kind kind_;
};
class L1ErrorMinimization : public NormMinimization {
public:
L1ErrorMinimization() : NormMinimization(L1) {}
};
class P99 : public QuantizationErrorMinimization {
public:
float threshold_;
P99(float p99_threshold = 0.99) : threshold_(p99_threshold) {}
TensorQuantizationParams ChooseQuantizationParams(
const Histogram& hist,
bool preserve_sparsity = true,
int precision = 8) override;
}; // class P99QuantizationFactory
} // namespace dnnlowp
|
