Xenova HF staff commited on
Commit
6f7830a
1 Parent(s): d8b94eb

Upload quant_config.json

Browse files
Files changed (1) hide show
  1. quant_config.json +74 -48
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
- "Mul",
8
- "Concat",
9
- "Sub",
10
- "ReduceMean",
11
- "Softmax",
12
- "Transpose",
13
  "Div",
14
- "Pow",
15
  "Gather",
16
- "Constant",
17
- "Conv",
18
- "Erf",
19
  "MatMul",
20
- "Unsqueeze",
21
- "Sqrt",
22
  "Add",
 
 
 
 
 
 
 
 
 
 
 
 
 
23
  "Shape",
24
- "Reshape"
 
 
 
25
  ],
26
- "weight_type": "QUInt8"
27
  },
28
- "decoder_model": {
29
  "op_types": [
30
- "Range",
31
- "ConstantOfShape",
32
- "MatMul",
33
- "Sqrt",
34
- "Mul",
35
  "Concat",
36
  "Less",
37
  "ReduceMean",
38
- "Sub",
39
- "Pow",
40
  "Erf",
 
 
 
 
 
 
41
  "Slice",
 
42
  "Cast",
43
- "Gather",
44
  "Where",
45
- "Softmax",
46
  "Constant",
47
- "Squeeze",
48
- "Add",
49
  "Shape",
50
  "Reshape",
 
51
  "Transpose",
52
- "Div",
53
- "Unsqueeze",
54
- "Expand",
55
- "Equal"
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
- "Mul",
 
68
  "Concat",
 
 
69
  "Sub",
70
- "ReduceMean",
71
- "Softmax",
 
72
  "Transpose",
73
- "Div",
74
- "Pow",
75
- "Shape",
76
- "Constant",
77
- "Erf",
78
- "Slice",
79
  "MatMul",
80
- "Unsqueeze",
81
  "Sqrt",
82
- "Add",
83
- "Gather",
84
- "Reshape"
 
 
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
  }