-
Notifications
You must be signed in to change notification settings - Fork 0
/
samet_brevitas.py
139 lines (128 loc) · 5.41 KB
/
samet_brevitas.py
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
from torch import nn
import brevitas.nn as qnn
class AI85KWS20Netv3_brevitas(nn.Module):
"""
Compound KWS20 v3 Audio net, all with Conv1Ds, but with brevitas
"""
# num_classes = n keywords + 1 unknown
def __init__(
self,
num_classes=21,
num_channels=128,
dimensions=(128, 1), # pylint: disable=unused-argument
bias=False,
weight_quant_type='INT',
weight_bit_width=4,
weight_scaling_impl_type='CONST',
weight_scaling_const=1,
bias_quant_type='INT',
bias_bit_width=4,
act_quant_type='INT',
act_bit_width=4,
act_scaling_impl_type='CONST',
act_scaling_const=1,
**kwargs):
super().__init__()
self.indentity = qnn.QuantIdentity(bit_width=4,
return_quant_tensor=False)
self.voice_conv1 = qnn.QuantConv1d(num_channels, 100, 1, bias=False,
stride=1, padding=0,
weight_bit_width=4,
)
self.voice_conv1_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
# T: 128 F: 100
self.voice_conv2 = qnn.QuantConv1d(100, 96, 3, bias=False, stride=1,
padding=0, weight_bit_width=4,
)
self.voice_conv2_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
self.dropout1 = qnn.QuantDropout(p=0.5)
# T: 126 F : 96
self.voice_conv3 = qnn.QuantConv1d(96, 64, 3, bias=False, stride=1,
padding=0, weight_bit_width=4,
)
self.voice_conv3_maxpool = qnn.QuantMaxPool1d(kernel_size=2, stride=2)
self.voice_conv3_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
# T: 62 F : 64
self.voice_conv4 = qnn.QuantConv1d(64, 48, 3, bias=False, stride=1,
padding=0, weight_bit_width=4,
)
self.voice_conv4_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
self.dropout2 = qnn.QuantDropout(p=0.5)
# T : 60 F : 48
self.kws_conv1 = qnn.QuantConv1d(48, 64, 3, bias=False, stride=1,
padding=0, weight_bit_width=4,
)
self.kws_conv1_maxpool = qnn.QuantMaxPool1d(kernel_size=2, stride=2)
self.kws_conv1_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
# T: 30 F : 64
self.kws_conv2 = qnn.QuantConv1d(64, 96, 3, bias=False,
stride=1, padding=0,
weight_bit_width=4,
)
self.kws_conv2_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
self.dropout3 = qnn.QuantDropout(p=0.5)
# T: 28 F : 96
self.kws_conv3 = qnn.QuantConv1d(96, 100, 3, bias=False,
stride=1, padding=0,
weight_bit_width=4,
)
self.kws_conv3_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
# T : 14 F: 100
self.kws_conv4 = qnn.QuantConv1d(100, 64, 6, bias=False,
stride=1, padding=0,
weight_bit_width=4)
self.kws_conv4_maxpool = qnn.QuantMaxPool1d(kernel_size=2, stride=2)
self.kws_conv4_relu = qnn.QuantReLU(bit_width=4,
return_quant_tensor=False)
# T : 2 F: 128
self.fc = nn.Linear(640, num_classes, bias=False)
def forward(self, x): # pylint: disable=arguments-differ
"""Forward prop"""
# Run CNN
x = self.indentity(x)
x = self.voice_conv1(x)
x = self.voice_conv1_relu(x)
x = self.voice_conv2(x)
x = self.voice_conv2_relu(x)
x = self.dropout1(x)
x = self.voice_conv3(x)
x = self.voice_conv3_maxpool(x)
x = self.voice_conv3_relu(x)
x = self.voice_conv4(x)
x = self.voice_conv4_relu(x)
x = self.dropout2(x)
x = self.kws_conv1(x)
x = self.kws_conv1_maxpool(x)
x = self.kws_conv1_relu(x)
x = self.kws_conv2(x)
x = self.dropout3(x)
x = self.kws_conv2_relu(x)
x = self.kws_conv3(x)
x = self.kws_conv3_relu(x)
x = self.kws_conv4(x)
x = self.kws_conv4_maxpool(x)
x = self.kws_conv4_relu(x)
x = x.view(x.size(0), -1)
x = self.fc(x)
return x
def samet_brevitas(pretrained=False, **kwargs):
"""
Constructs a AI85KWS20Net model.
rn AI85KWS20Net(**kwargs)
"""
assert not pretrained
return AI85KWS20Netv3_brevitas(**kwargs)
models = [
{
'name': 'samet_brevitas',
'min_input': 1,
'dim': 1,
},
]