Xenova HF staff commited on
Commit
a27778a
1 Parent(s): 4e6668f

Upload quant_config.json

Browse files
Files changed (1) hide show
  1. quant_config.json +75 -49
quant_config.json CHANGED
@@ -2,88 +2,114 @@
2
  "per_channel": false,
3
  "reduce_range": false,
4
  "per_model_config": {
5
- "encoder_model": {
6
  "op_types": [
7
  "Div",
8
- "Erf",
9
  "Gather",
10
- "MatMul",
11
- "Conv",
12
  "ReduceMean",
13
- "Pow",
14
- "Softmax",
 
 
15
  "Sub",
 
 
 
 
 
 
 
 
16
  "Constant",
17
- "Add",
18
- "Transpose",
19
- "Mul",
20
- "Unsqueeze",
21
  "Shape",
22
- "Sqrt",
23
  "Reshape",
24
- "Concat"
 
 
25
  ],
26
- "weight_type": "QUInt8"
27
  },
28
- "decoder_model": {
29
  "op_types": [
30
- "Cast",
31
- "Squeeze",
32
- "Slice",
33
- "Less",
34
- "Sub",
35
  "Div",
36
- "Expand",
37
  "Gather",
38
- "Range",
39
- "ConstantOfShape",
40
- "Mul",
41
- "Shape",
42
- "Sqrt",
43
- "Reshape",
44
- "ReduceMean",
45
- "Pow",
46
- "Constant",
47
- "Add",
48
- "Unsqueeze",
49
  "Concat",
50
- "Equal",
 
51
  "MatMul",
52
  "Erf",
53
- "Softmax",
 
 
 
 
 
 
 
 
 
54
  "Where",
55
- "Transpose"
 
 
 
 
 
 
 
 
56
  ],
57
  "weight_type": "QInt8"
58
  },
59
- "decoder_model_merged": {
60
  "op_types": [
61
- "If"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
62
  ],
63
  "weight_type": "QInt8"
64
  },
65
- "decoder_with_past_model": {
66
  "op_types": [
67
  "Div",
68
  "Gather",
69
- "MatMul",
70
- "Erf",
71
- "ReduceMean",
72
  "Pow",
73
- "Softmax",
74
- "Slice",
75
  "Sub",
76
- "Constant",
77
- "Add",
78
- "Transpose",
79
- "Mul",
80
  "Unsqueeze",
81
- "Shape",
 
 
 
 
82
  "Sqrt",
 
83
  "Reshape",
84
- "Concat"
 
 
85
  ],
86
- "weight_type": "QInt8"
87
  }
88
  }
89
  }
 
2
  "per_channel": false,
3
  "reduce_range": false,
4
  "per_model_config": {
5
+ "decoder_model": {
6
  "op_types": [
7
  "Div",
8
+ "Concat",
9
  "Gather",
10
+ "Less",
 
11
  "ReduceMean",
12
+ "MatMul",
13
+ "Erf",
14
+ "Add",
15
+ "Expand",
16
  "Sub",
17
+ "ConstantOfShape",
18
+ "Range",
19
+ "Sqrt",
20
+ "Slice",
21
+ "Squeeze",
22
+ "Cast",
23
+ "Equal",
24
+ "Where",
25
  "Constant",
26
+ "Pow",
27
+ "Softmax",
 
 
28
  "Shape",
 
29
  "Reshape",
30
+ "Mul",
31
+ "Transpose",
32
+ "Unsqueeze"
33
  ],
34
+ "weight_type": "QInt8"
35
  },
36
+ "decoder_model_merged": {
37
  "op_types": [
 
 
 
 
 
38
  "Div",
 
39
  "Gather",
 
 
 
 
 
 
 
 
 
 
 
40
  "Concat",
41
+ "Less",
42
+ "ReduceMean",
43
  "MatMul",
44
  "Erf",
45
+ "Add",
46
+ "Expand",
47
+ "Sub",
48
+ "ConstantOfShape",
49
+ "Range",
50
+ "Sqrt",
51
+ "Slice",
52
+ "Squeeze",
53
+ "Cast",
54
+ "Equal",
55
  "Where",
56
+ "If",
57
+ "Constant",
58
+ "Pow",
59
+ "Softmax",
60
+ "Shape",
61
+ "Reshape",
62
+ "Mul",
63
+ "Transpose",
64
+ "Unsqueeze"
65
  ],
66
  "weight_type": "QInt8"
67
  },
68
+ "decoder_with_past_model": {
69
  "op_types": [
70
+ "Div",
71
+ "Concat",
72
+ "Gather",
73
+ "Constant",
74
+ "Pow",
75
+ "Sub",
76
+ "Mul",
77
+ "Transpose",
78
+ "Add",
79
+ "MatMul",
80
+ "Softmax",
81
+ "Sqrt",
82
+ "Slice",
83
+ "Shape",
84
+ "Reshape",
85
+ "ReduceMean",
86
+ "Erf",
87
+ "Unsqueeze"
88
  ],
89
  "weight_type": "QInt8"
90
  },
91
+ "encoder_model": {
92
  "op_types": [
93
  "Div",
94
  "Gather",
95
+ "Concat",
96
+ "Constant",
 
97
  "Pow",
 
 
98
  "Sub",
 
 
 
 
99
  "Unsqueeze",
100
+ "Mul",
101
+ "Conv",
102
+ "Transpose",
103
+ "MatMul",
104
+ "Softmax",
105
  "Sqrt",
106
+ "Shape",
107
  "Reshape",
108
+ "ReduceMean",
109
+ "Erf",
110
+ "Add"
111
  ],
112
+ "weight_type": "QUInt8"
113
  }
114
  }
115
  }