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
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
|
from functools import partial
import caffe2.python.hypothesis_test_util as hu
import numpy as np
from caffe2.python import core
def ref_adagrad(
param_in,
mom_in,
grad,
lr,
epsilon,
using_fp16=False,
output_effective_lr=False,
output_effective_lr_and_update=False,
decay=1.0,
row_wise=False,
weight_decay=0.0,
counter_halflife=-1,
count=None, # only used when counter_halflife != -1
):
mom_in_f32 = mom_in
param_in_f32 = param_in
if using_fp16:
mom_in_f32 = mom_in.astype(np.float32)
param_in_f32 = param_in.astype(np.float32)
if count and count > 0 and counter_halflife > 0:
weight_decay *= counter_halflife / count
grad_temp = grad + weight_decay * param_in_f32
if row_wise:
mom_out = decay * mom_in_f32 + np.mean(np.square(grad_temp))
else:
mom_out = decay * mom_in_f32 + np.square(grad_temp)
effective_lr = lr / (np.sqrt(mom_out) + epsilon)
grad_adj = effective_lr * grad_temp
param_out = param_in_f32 + grad_adj
if output_effective_lr_and_update:
if using_fp16:
return (
param_out.astype(np.float16),
mom_out.astype(np.float16),
effective_lr.astype(np.float16),
grad_adj.astype(np.float16),
)
else:
return (
param_out.astype(np.float32),
mom_out.astype(np.float32),
effective_lr.astype(np.float32),
grad_adj.astype(np.float32),
)
elif output_effective_lr:
if using_fp16:
return (
param_out.astype(np.float16),
mom_out.astype(np.float16),
effective_lr.astype(np.float16),
)
else:
return (
param_out.astype(np.float32),
mom_out.astype(np.float32),
effective_lr.astype(np.float32),
)
if using_fp16:
return (param_out.astype(np.float16), mom_out.astype(np.float16))
else:
return (param_out.astype(np.float32), mom_out.astype(np.float32))
def adagrad_sparse_test_helper(
parent_test,
inputs,
lr,
epsilon,
engine,
ref_adagrad,
gc,
dc,
row_wise=False,
weight_decay=0.0,
counter_halflife=-1,
):
param, momentum, grad = inputs
if row_wise:
# For row-wise adagrad, only take the first element of each row
momentum = momentum.reshape(momentum.shape[0], -1)[:, 0]
momentum = np.abs(momentum)
lr = np.array([lr], dtype=np.float32)
count = None
if counter_halflife != -1:
count = np.random.rand(param.shape[0])
# Create an indexing array containing values that are lists of indices,
# which index into grad
if grad.size == 0:
indices = np.empty(shape=(0,), dtype=np.int)
else:
indices = np.random.choice(
np.arange(grad.shape[0]),
size=np.random.randint(grad.shape[0]),
replace=False,
)
# Sparsify grad
grad = grad[indices]
op = core.CreateOperator(
"RowWiseSparseAdagrad" if row_wise else "SparseAdagrad",
["param", "momentum", "indices", "grad", "lr"] if count is None else ["param", "momentum", "indices", "grad", "lr", "count"],
["param", "momentum"],
epsilon=epsilon,
weight_decay=weight_decay,
counter_halflife=counter_halflife,
engine=engine,
device_option=gc,
)
def ref_sparse(param, momentum, indices, grad, lr, count=None, ref_using_fp16=False):
param_out = np.copy(param)
momentum_out = np.copy(momentum)
# Need to do this because it's possible ref_adagrad's using_fp16 could
# have been already specialized.
ref_adagrad_temp = (
partial(ref_adagrad, using_fp16=ref_using_fp16)
if ref_using_fp16
else ref_adagrad
)
for i, index in enumerate(indices):
param_out[index], momentum_out[index] = ref_adagrad_temp(
param[index],
momentum[index],
grad[i],
lr,
epsilon,
weight_decay=weight_decay,
counter_halflife=counter_halflife,
count=None if count is None else count[index],
)
return (param_out, momentum_out)
ref_using_fp16_values = [False]
if gc == hu.gpu_do and not row_wise:
ref_using_fp16_values.append(True)
for ref_using_fp16 in ref_using_fp16_values:
if ref_using_fp16:
print("test_sparse_adagrad with half precision embedding")
momentum_i = momentum.astype(np.float16)
param_i = param.astype(np.float16)
else:
print("test_sparse_adagrad with full precision embedding")
momentum_i = momentum.astype(np.float32)
param_i = param.astype(np.float32)
parent_test.assertReferenceChecks(
gc,
op,
[param_i, momentum_i, indices, grad, lr, count, ref_using_fp16],
ref_sparse
)
|
