msluszniak commited on
Commit
6daac7d
·
verified ·
1 Parent(s): 8589396

Add spec-compliant config.json files

Browse files
Files changed (1) hide show
  1. xnnpack/config.json +130 -2
xnnpack/config.json CHANGED
@@ -12,13 +12,141 @@
12
  "file": "phi_4_mini_xnnpack_bf16.pte",
13
  "precision": "bf16",
14
  "quantized": false,
15
- "default": true
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
16
  },
17
  {
18
  "file": "phi_4_mini_xnnpack_8da4w.pte",
19
  "precision": "8da4w",
20
  "quantized": true,
21
- "default": true
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22
  }
23
  ]
24
  }
 
12
  "file": "phi_4_mini_xnnpack_bf16.pte",
13
  "precision": "bf16",
14
  "quantized": false,
15
+ "default": true,
16
+ "methods": {
17
+ "enable_dynamic_shape": {
18
+ "inputs": [],
19
+ "outputs": []
20
+ },
21
+ "forward": {
22
+ "inputs": [
23
+ {
24
+ "shape": [
25
+ 1,
26
+ 127
27
+ ],
28
+ "dtype": "int64"
29
+ },
30
+ {
31
+ "shape": [
32
+ 1
33
+ ],
34
+ "dtype": "int64"
35
+ }
36
+ ],
37
+ "outputs": [
38
+ {
39
+ "shape": [
40
+ 1,
41
+ 200064
42
+ ],
43
+ "dtype": "bfloat16"
44
+ }
45
+ ]
46
+ },
47
+ "get_bos_id": {
48
+ "inputs": [],
49
+ "outputs": []
50
+ },
51
+ "get_eos_ids": {
52
+ "inputs": [],
53
+ "outputs": []
54
+ },
55
+ "get_max_context_len": {
56
+ "inputs": [],
57
+ "outputs": []
58
+ },
59
+ "get_max_seq_len": {
60
+ "inputs": [],
61
+ "outputs": []
62
+ },
63
+ "get_n_layers": {
64
+ "inputs": [],
65
+ "outputs": []
66
+ },
67
+ "get_vocab_size": {
68
+ "inputs": [],
69
+ "outputs": []
70
+ },
71
+ "use_kv_cache": {
72
+ "inputs": [],
73
+ "outputs": []
74
+ },
75
+ "use_sdpa_with_kv_cache": {
76
+ "inputs": [],
77
+ "outputs": []
78
+ }
79
+ }
80
  },
81
  {
82
  "file": "phi_4_mini_xnnpack_8da4w.pte",
83
  "precision": "8da4w",
84
  "quantized": true,
85
+ "default": true,
86
+ "methods": {
87
+ "enable_dynamic_shape": {
88
+ "inputs": [],
89
+ "outputs": []
90
+ },
91
+ "forward": {
92
+ "inputs": [
93
+ {
94
+ "shape": [
95
+ 1,
96
+ 2047
97
+ ],
98
+ "dtype": "int64"
99
+ },
100
+ {
101
+ "shape": [
102
+ 1
103
+ ],
104
+ "dtype": "int64"
105
+ }
106
+ ],
107
+ "outputs": [
108
+ {
109
+ "shape": [
110
+ 1,
111
+ 200064
112
+ ],
113
+ "dtype": "float32"
114
+ }
115
+ ]
116
+ },
117
+ "get_bos_id": {
118
+ "inputs": [],
119
+ "outputs": []
120
+ },
121
+ "get_eos_ids": {
122
+ "inputs": [],
123
+ "outputs": []
124
+ },
125
+ "get_max_context_len": {
126
+ "inputs": [],
127
+ "outputs": []
128
+ },
129
+ "get_max_seq_len": {
130
+ "inputs": [],
131
+ "outputs": []
132
+ },
133
+ "get_n_layers": {
134
+ "inputs": [],
135
+ "outputs": []
136
+ },
137
+ "get_vocab_size": {
138
+ "inputs": [],
139
+ "outputs": []
140
+ },
141
+ "use_kv_cache": {
142
+ "inputs": [],
143
+ "outputs": []
144
+ },
145
+ "use_sdpa_with_kv_cache": {
146
+ "inputs": [],
147
+ "outputs": []
148
+ }
149
+ }
150
  }
151
  ]
152
  }