Xenova HF staff commited on
Commit
27103eb
1 Parent(s): d0cc1bf

Update quantize_config.json

Browse files
Files changed (1) hide show
  1. quantize_config.json +102 -28
quantize_config.json CHANGED
@@ -1,31 +1,105 @@
1
  {
2
- "per_channel": true,
3
- "reduce_range": true,
4
- "per_model_config": {
5
- "model": {
6
- "op_types": [
7
- "Unsqueeze",
8
- "Softmax",
9
- "Add",
10
- "Mul",
11
- "Reshape",
12
- "Pow",
13
- "InstanceNormalization",
14
- "Shape",
15
- "Constant",
16
- "MatMul",
17
- "ReduceMean",
18
- "Erf",
19
- "Sqrt",
20
- "Div",
21
- "Gather",
22
- "Concat",
23
- "Slice",
24
- "Sub",
25
- "Conv",
26
- "Transpose"
27
- ],
28
- "weight_type": "QUInt8"
29
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30
  }
31
  }
 
1
  {
2
+ "fp16": {},
3
+ "q8": {
4
+ "per_model_config": {
5
+ "model": {
6
+ "op_types": [
7
+ "Add",
8
+ "Concat",
9
+ "Constant",
10
+ "Conv",
11
+ "Div",
12
+ "Erf",
13
+ "Gather",
14
+ "InstanceNormalization",
15
+ "MatMul",
16
+ "Mul",
17
+ "Pow",
18
+ "ReduceMean",
19
+ "Reshape",
20
+ "Shape",
21
+ "Slice",
22
+ "Softmax",
23
+ "Sqrt",
24
+ "Sub",
25
+ "Transpose",
26
+ "Unsqueeze"
27
+ ],
28
+ "weight_type": "QUInt8"
29
+ }
30
+ },
31
+ "per_channel": false,
32
+ "reduce_range": false
33
+ },
34
+ "int8": {
35
+ "per_model_config": {
36
+ "model": {
37
+ "op_types": [
38
+ "Add",
39
+ "Concat",
40
+ "Constant",
41
+ "Conv",
42
+ "Div",
43
+ "Erf",
44
+ "Gather",
45
+ "InstanceNormalization",
46
+ "MatMul",
47
+ "Mul",
48
+ "Pow",
49
+ "ReduceMean",
50
+ "Reshape",
51
+ "Shape",
52
+ "Slice",
53
+ "Softmax",
54
+ "Sqrt",
55
+ "Sub",
56
+ "Transpose",
57
+ "Unsqueeze"
58
+ ],
59
+ "weight_type": "QInt8"
60
+ }
61
+ },
62
+ "per_channel": false,
63
+ "reduce_range": false
64
+ },
65
+ "uint8": {
66
+ "per_model_config": {
67
+ "model": {
68
+ "op_types": [
69
+ "Add",
70
+ "Concat",
71
+ "Constant",
72
+ "Conv",
73
+ "Div",
74
+ "Erf",
75
+ "Gather",
76
+ "InstanceNormalization",
77
+ "MatMul",
78
+ "Mul",
79
+ "Pow",
80
+ "ReduceMean",
81
+ "Reshape",
82
+ "Shape",
83
+ "Slice",
84
+ "Softmax",
85
+ "Sqrt",
86
+ "Sub",
87
+ "Transpose",
88
+ "Unsqueeze"
89
+ ],
90
+ "weight_type": "QUInt8"
91
+ }
92
+ },
93
+ "per_channel": false,
94
+ "reduce_range": false
95
+ },
96
+ "q4": {
97
+ "block_size": 32,
98
+ "is_symmetric": true,
99
+ "accuracy_level": null
100
+ },
101
+ "bnb4": {
102
+ "block_size": 64,
103
+ "quant_type": 1
104
  }
105
  }