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
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
|
#ifndef CAFFE2_OPERATORS_RMS_NORM_OP_H_
#define CAFFE2_OPERATORS_RMS_NORM_OP_H_
#include "caffe2/core/context.h"
#include "caffe2/core/operator.h"
#include "caffe2/core/types.h"
#include "caffe2/utils/math.h"
namespace caffe2 {
// RMSNorm op.
// https://openreview.net/pdf?id=SygkZ3MTJE
template <class Context>
class RMSNormOp final : public Operator<Context> {
public:
USE_OPERATOR_CONTEXT_FUNCTIONS;
template <class... Args>
explicit RMSNormOp(Args&&... args)
: Operator<Context>(std::forward<Args>(args)...),
OP_SINGLE_ARG(int, "axis", axis_, 1),
OP_SINGLE_ARG(float, "eps", eps_, 0.0f) {}
bool RunOnDevice() override {
return DispatchHelper<TensorTypes<float, double>>::call(this, Input(0));
}
template <typename T>
bool DoRunWithType();
private:
const int axis_;
const float eps_;
};
template <class Context>
class RMSNormGradientOp final : public Operator<Context> {
public:
USE_OPERATOR_CONTEXT_FUNCTIONS;
template <class... Args>
explicit RMSNormGradientOp(Args&&... args)
: Operator<Context>(std::forward<Args>(args)...),
OP_SINGLE_ARG(int, "axis", axis_, 1) {}
bool RunOnDevice() override {
return DispatchHelper<TensorTypes<float, double>>::call(this, Input(0));
}
template <typename T>
bool DoRunWithType() {
const auto& dY = Input(0);
const auto& X = Input(1);
const auto& gamma = Input(2);
const auto& rrms = Input(3);
const int canonical_axis = X.canonical_axis_index(axis_);
const int64_t M = X.size_to_dim(canonical_axis);
const int64_t N = X.size_from_dim(canonical_axis);
auto* dX = Output(0, X.sizes(), at::dtype<T>());
auto* dgamma = Output(1, gamma.sizes(), at::dtype<T>());
auto* dbeta = Output(2, gamma.sizes(), at::dtype<T>());
const T* dY_data = dY.template data<T>();
const T* X_data = X.template data<T>();
const T* gamma_data = gamma.template data<T>();
const T* rrms_data = rrms.template data<T>();
T* dX_data = dX->template mutable_data<T>();
T* dgamma_data = dgamma->template mutable_data<T>();
T* dbeta_data = dbeta->template mutable_data<T>();
if (M == 0) {
math::Set<T, Context>(N, T(0), dgamma_data, &context_);
math::Set<T, Context>(N, T(0), dbeta_data, &context_);
return true;
}
RMSNormBackward<T>(M, N, dY_data, X_data, gamma_data, rrms_data, dX_data);
GammaBetaBackward<T>(
M, N, dY_data, X_data, rrms_data, dgamma_data, dbeta_data);
return true;
}
private:
template <typename T>
void RMSNormBackward(
int64_t M,
int64_t N,
const T* dY,
const T* X,
const T* gamma,
const T* rrms,
T* dX);
template <typename T>
void GammaBetaBackward(
int64_t M,
int64_t N,
const T* dY,
const T* X,
const T* rrms,
T* dgamma,
T* dbeta);
const int axis_;
Tensor c2_;
};
} // namespace caffe2
#endif // CAFFE2_OPERATORS_RMS_NORM_OP_H_
|
