Hugging Face's logo

software-mansion
/
react-native-executorch-efficientnet-v2-s

Image Classification
ExecuTorch
Model card Files
xet
 Community
msluszniak commited on
Commit
ceb6a91
·
1 Parent(s): f3b9614

Add QNN a16w8 exports, one per Hexagon version (v69-v81)

Browse files

Only v81 is device-validated (Galaxy S26 Ultra, SM8850: 74.1% top-1 on ImageNetV2, 2.4 ms). v69-v79 are the same recipe, untested.

Files changed (6) hide show
  1. qnn/config.json +152 -0
  2. qnn/efficientnet_v2_s_qnn_a16w8_v69.pte +3 -0
  3. qnn/efficientnet_v2_s_qnn_a16w8_v73.pte +3 -0
  4. qnn/efficientnet_v2_s_qnn_a16w8_v75.pte +3 -0
  5. qnn/efficientnet_v2_s_qnn_a16w8_v79.pte +3 -0
  6. qnn/efficientnet_v2_s_qnn_a16w8_v81.pte +3 -0
qnn/config.json ADDED
@@ -0,0 +1,152 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
3
+ "model": "efficientnet_v2_s",
4
+ "family": "efficientnet",
5
+ "capabilities": [
6
+ "classification"
7
+ ],
8
+ "backend": "qnn",
9
+ "license": "bsd-3-clause",
10
+ "variants": [
11
+ {
12
+ "file": "efficientnet_v2_s_qnn_a16w8_v69.pte",
13
+ "precision": "a16w8",
14
+ "methods": {
15
+ "forward": {
16
+ "inputs": [
17
+ {
18
+ "shape": [
19
+ 1,
20
+ 3,
21
+ 384,
22
+ 384
23
+ ],
24
+ "dtype": "float32"
25
+ }
26
+ ],
27
+ "outputs": [
28
+ {
29
+ "shape": [
30
+ 1,
31
+ 1000
32
+ ],
33
+ "dtype": "float32"
34
+ }
35
+ ]
36
+ }
37
+ }
38
+ },
39
+ {
40
+ "file": "efficientnet_v2_s_qnn_a16w8_v73.pte",
41
+ "precision": "a16w8",
42
+ "methods": {
43
+ "forward": {
44
+ "inputs": [
45
+ {
46
+ "shape": [
47
+ 1,
48
+ 3,
49
+ 384,
50
+ 384
51
+ ],
52
+ "dtype": "float32"
53
+ }
54
+ ],
55
+ "outputs": [
56
+ {
57
+ "shape": [
58
+ 1,
59
+ 1000
60
+ ],
61
+ "dtype": "float32"
62
+ }
63
+ ]
64
+ }
65
+ }
66
+ },
67
+ {
68
+ "file": "efficientnet_v2_s_qnn_a16w8_v75.pte",
69
+ "precision": "a16w8",
70
+ "methods": {
71
+ "forward": {
72
+ "inputs": [
73
+ {
74
+ "shape": [
75
+ 1,
76
+ 3,
77
+ 384,
78
+ 384
79
+ ],
80
+ "dtype": "float32"
81
+ }
82
+ ],
83
+ "outputs": [
84
+ {
85
+ "shape": [
86
+ 1,
87
+ 1000
88
+ ],
89
+ "dtype": "float32"
90
+ }
91
+ ]
92
+ }
93
+ }
94
+ },
95
+ {
96
+ "file": "efficientnet_v2_s_qnn_a16w8_v79.pte",
97
+ "precision": "a16w8",
98
+ "methods": {
99
+ "forward": {
100
+ "inputs": [
101
+ {
102
+ "shape": [
103
+ 1,
104
+ 3,
105
+ 384,
106
+ 384
107
+ ],
108
+ "dtype": "float32"
109
+ }
110
+ ],
111
+ "outputs": [
112
+ {
113
+ "shape": [
114
+ 1,
115
+ 1000
116
+ ],
117
+ "dtype": "float32"
118
+ }
119
+ ]
120
+ }
121
+ }
122
+ },
123
+ {
124
+ "file": "efficientnet_v2_s_qnn_a16w8_v81.pte",
125
+ "precision": "a16w8",
126
+ "methods": {
127
+ "forward": {
128
+ "inputs": [
129
+ {
130
+ "shape": [
131
+ 1,
132
+ 3,
133
+ 384,
134
+ 384
135
+ ],
136
+ "dtype": "float32"
137
+ }
138
+ ],
139
+ "outputs": [
140
+ {
141
+ "shape": [
142
+ 1,
143
+ 1000
144
+ ],
145
+ "dtype": "float32"
146
+ }
147
+ ]
148
+ }
149
+ }
150
+ }
151
+ ]
152
+ }
qnn/efficientnet_v2_s_qnn_a16w8_v69.pte ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d449e33e881d41f18ffc132c9c689f890141aa77742c32b7b8e918645b5b7ea4
3
+ size 32839168
qnn/efficientnet_v2_s_qnn_a16w8_v73.pte ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac2c1e554c2f178051a96d07978520ef5165b1f0fcf67e29940de92d21f3d700
3
+ size 25556480
qnn/efficientnet_v2_s_qnn_a16w8_v75.pte ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fd0e960b3547f7e1bac16e4ed59b72d117ffcb235b19d5fc79ac2dc97c4ec8a0
3
+ size 25556480
qnn/efficientnet_v2_s_qnn_a16w8_v79.pte ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:36cd725ef001a4cbee99120db0b271512e593383ef67f4d3236eb5b67af12680
3
+ size 25646592
qnn/efficientnet_v2_s_qnn_a16w8_v81.pte ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f192c8acb92a63d6403149d90c0288ee1d34fd9f4b80a46b3cc90d1bca0b4d80
3
+ size 25712128