Upload 15 files
Browse files- onnx/decoder_model.onnx +1 -1
- onnx/decoder_model_merged.onnx +1 -1
- onnx/decoder_model_merged_quantized.onnx +1 -1
- onnx/decoder_model_quantized.onnx +1 -1
- onnx/decoder_with_past_model.onnx +1 -1
- onnx/decoder_with_past_model_quantized.onnx +1 -1
- onnx/encoder_model.onnx +1 -1
- onnx/encoder_model_quantized.onnx +2 -2
- quantize_config.json +80 -80
onnx/decoder_model.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 166670441
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:f02dbc32615cb67509fc56a8b632cd089ce83763c4fdd0684a0db2ea75d6d92e
|
3 |
size 166670441
|
onnx/decoder_model_merged.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 166832634
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e7f873f936a3df005ec66a0bdb59ec46867abee46299d851931ac58a8e50ff51
|
3 |
size 166832634
|
onnx/decoder_model_merged_quantized.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 42431444
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:360c22e8d266e3c78e33f8a6fd84aeb6888bd4f6bc0f89bc4f6e3313a97f00a5
|
3 |
size 42431444
|
onnx/decoder_model_quantized.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 42154677
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b7f3596e68f9fbd78d88a2d74b31781feb01f9d3a815f058ab0e36434f8c9fd1
|
3 |
size 42154677
|
onnx/decoder_with_past_model.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 154076576
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:b661a1113f4ff1449cf633aa8f68c15ddb16b6d2df57e898b59a9b5f40fa00cf
|
3 |
size 154076576
|
onnx/decoder_with_past_model_quantized.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 38952139
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3608448e65af845b3a34d6fbaed226ffc4f8b774cf969777a024929636a6fe11
|
3 |
size 38952139
|
onnx/encoder_model.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 141404308
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d2816dcb1b5df89b4c5099288652d25850ccf10d2bce626fe29166c02c4da8ec
|
3 |
size 141404308
|
onnx/encoder_model_quantized.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:2615f0642add5ca2a89c7a748823ead83c9ef895600d312a25f850c435e60326
|
3 |
+
size 35640025
|
quantize_config.json
CHANGED
@@ -4,134 +4,134 @@
|
|
4 |
"per_model_config": {
|
5 |
"decoder_model": {
|
6 |
"op_types": [
|
7 |
-
"
|
8 |
-
"
|
9 |
-
"ReduceMean",
|
10 |
-
"Cast",
|
11 |
-
"Min",
|
12 |
-
"Expand",
|
13 |
-
"Shape",
|
14 |
"Softmax",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
"Range",
|
16 |
-
"Pow",
|
17 |
"Gather",
|
|
|
|
|
|
|
|
|
|
|
18 |
"Sqrt",
|
19 |
-
"
|
|
|
|
|
20 |
"Add",
|
21 |
"Relu",
|
22 |
-
"
|
23 |
-
"
|
24 |
-
"ConstantOfShape",
|
25 |
-
"Sub",
|
26 |
-
"Concat",
|
27 |
-
"Mul",
|
28 |
-
"Div",
|
29 |
-
"Tile",
|
30 |
"Neg",
|
31 |
-
"Unsqueeze",
|
32 |
"Constant",
|
33 |
-
"
|
34 |
-
"Where"
|
35 |
],
|
36 |
"weight_type": "QInt8"
|
37 |
},
|
38 |
"decoder_model_merged": {
|
39 |
"op_types": [
|
40 |
-
"
|
41 |
-
"
|
42 |
-
"
|
43 |
-
"
|
44 |
"Min",
|
|
|
|
|
|
|
|
|
45 |
"Slice",
|
46 |
-
"
|
47 |
-
"Shape",
|
48 |
-
"Softmax",
|
49 |
"Range",
|
50 |
-
"Pow",
|
51 |
"Gather",
|
|
|
|
|
|
|
|
|
|
|
52 |
"Sqrt",
|
53 |
-
"
|
54 |
-
"
|
55 |
-
"Relu",
|
56 |
"Log",
|
57 |
-
"MatMul",
|
58 |
"If",
|
59 |
-
"
|
60 |
-
"
|
61 |
-
"
|
62 |
-
"
|
63 |
-
"Div",
|
64 |
-
"Tile",
|
65 |
"Neg",
|
66 |
-
"Unsqueeze",
|
67 |
"Constant",
|
68 |
-
"
|
69 |
-
"Where"
|
70 |
],
|
71 |
"weight_type": "QInt8"
|
72 |
},
|
73 |
"decoder_with_past_model": {
|
74 |
"op_types": [
|
75 |
-
"
|
76 |
-
"
|
77 |
-
"
|
78 |
-
"
|
79 |
"Min",
|
|
|
|
|
|
|
|
|
80 |
"Slice",
|
81 |
-
"
|
82 |
-
"Shape",
|
83 |
-
"Softmax",
|
84 |
"Range",
|
85 |
-
"Pow",
|
86 |
"Gather",
|
|
|
|
|
|
|
|
|
|
|
87 |
"Sqrt",
|
88 |
-
"
|
|
|
|
|
89 |
"Add",
|
90 |
"Relu",
|
91 |
-
"
|
92 |
-
"
|
93 |
-
"ConstantOfShape",
|
94 |
-
"Sub",
|
95 |
-
"Concat",
|
96 |
-
"Mul",
|
97 |
-
"Div",
|
98 |
-
"Tile",
|
99 |
"Neg",
|
100 |
-
"Unsqueeze",
|
101 |
"Constant",
|
102 |
-
"
|
103 |
-
"Where"
|
104 |
],
|
105 |
"weight_type": "QInt8"
|
106 |
},
|
107 |
"encoder_model": {
|
108 |
"op_types": [
|
109 |
-
"
|
110 |
-
"
|
111 |
-
"Cast",
|
112 |
-
"Min",
|
113 |
-
"Greater",
|
114 |
-
"Shape",
|
115 |
"Softmax",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
116 |
"Range",
|
117 |
-
"Pow",
|
118 |
"Gather",
|
|
|
|
|
|
|
119 |
"Sqrt",
|
120 |
-
"
|
|
|
|
|
|
|
121 |
"Add",
|
122 |
"Relu",
|
123 |
-
"
|
124 |
-
"Abs",
|
125 |
-
"MatMul",
|
126 |
-
"ConstantOfShape",
|
127 |
-
"Sub",
|
128 |
-
"Concat",
|
129 |
-
"Mul",
|
130 |
-
"Div",
|
131 |
-
"Unsqueeze",
|
132 |
-
"Constant",
|
133 |
"Transpose",
|
134 |
-
"
|
|
|
135 |
],
|
136 |
"weight_type": "QInt8"
|
137 |
}
|
|
|
4 |
"per_model_config": {
|
5 |
"decoder_model": {
|
6 |
"op_types": [
|
7 |
+
"Mul",
|
8 |
+
"Pow",
|
|
|
|
|
|
|
|
|
|
|
9 |
"Softmax",
|
10 |
+
"Reshape",
|
11 |
+
"Min",
|
12 |
+
"Less",
|
13 |
+
"ConstantOfShape",
|
14 |
+
"Div",
|
15 |
+
"MatMul",
|
16 |
+
"Unsqueeze",
|
17 |
"Range",
|
|
|
18 |
"Gather",
|
19 |
+
"Shape",
|
20 |
+
"Tile",
|
21 |
+
"ReduceMean",
|
22 |
+
"LessOrEqual",
|
23 |
+
"Sub",
|
24 |
"Sqrt",
|
25 |
+
"Cast",
|
26 |
+
"Where",
|
27 |
+
"Log",
|
28 |
"Add",
|
29 |
"Relu",
|
30 |
+
"Transpose",
|
31 |
+
"Expand",
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
"Neg",
|
|
|
33 |
"Constant",
|
34 |
+
"Concat"
|
|
|
35 |
],
|
36 |
"weight_type": "QInt8"
|
37 |
},
|
38 |
"decoder_model_merged": {
|
39 |
"op_types": [
|
40 |
+
"Mul",
|
41 |
+
"Pow",
|
42 |
+
"Softmax",
|
43 |
+
"Reshape",
|
44 |
"Min",
|
45 |
+
"Less",
|
46 |
+
"ConstantOfShape",
|
47 |
+
"Div",
|
48 |
+
"MatMul",
|
49 |
"Slice",
|
50 |
+
"Unsqueeze",
|
|
|
|
|
51 |
"Range",
|
|
|
52 |
"Gather",
|
53 |
+
"Shape",
|
54 |
+
"Tile",
|
55 |
+
"ReduceMean",
|
56 |
+
"LessOrEqual",
|
57 |
+
"Sub",
|
58 |
"Sqrt",
|
59 |
+
"Cast",
|
60 |
+
"Where",
|
|
|
61 |
"Log",
|
|
|
62 |
"If",
|
63 |
+
"Add",
|
64 |
+
"Relu",
|
65 |
+
"Transpose",
|
66 |
+
"Expand",
|
|
|
|
|
67 |
"Neg",
|
|
|
68 |
"Constant",
|
69 |
+
"Concat"
|
|
|
70 |
],
|
71 |
"weight_type": "QInt8"
|
72 |
},
|
73 |
"decoder_with_past_model": {
|
74 |
"op_types": [
|
75 |
+
"Mul",
|
76 |
+
"Pow",
|
77 |
+
"Softmax",
|
78 |
+
"Reshape",
|
79 |
"Min",
|
80 |
+
"Less",
|
81 |
+
"ConstantOfShape",
|
82 |
+
"Div",
|
83 |
+
"MatMul",
|
84 |
"Slice",
|
85 |
+
"Unsqueeze",
|
|
|
|
|
86 |
"Range",
|
|
|
87 |
"Gather",
|
88 |
+
"Shape",
|
89 |
+
"Tile",
|
90 |
+
"ReduceMean",
|
91 |
+
"LessOrEqual",
|
92 |
+
"Sub",
|
93 |
"Sqrt",
|
94 |
+
"Cast",
|
95 |
+
"Where",
|
96 |
+
"Log",
|
97 |
"Add",
|
98 |
"Relu",
|
99 |
+
"Transpose",
|
100 |
+
"Expand",
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
"Neg",
|
|
|
102 |
"Constant",
|
103 |
+
"Concat"
|
|
|
104 |
],
|
105 |
"weight_type": "QInt8"
|
106 |
},
|
107 |
"encoder_model": {
|
108 |
"op_types": [
|
109 |
+
"Mul",
|
110 |
+
"Pow",
|
|
|
|
|
|
|
|
|
111 |
"Softmax",
|
112 |
+
"Reshape",
|
113 |
+
"Min",
|
114 |
+
"Less",
|
115 |
+
"ConstantOfShape",
|
116 |
+
"Div",
|
117 |
+
"MatMul",
|
118 |
+
"Unsqueeze",
|
119 |
"Range",
|
|
|
120 |
"Gather",
|
121 |
+
"Shape",
|
122 |
+
"ReduceMean",
|
123 |
+
"Sub",
|
124 |
"Sqrt",
|
125 |
+
"Abs",
|
126 |
+
"Cast",
|
127 |
+
"Where",
|
128 |
+
"Log",
|
129 |
"Add",
|
130 |
"Relu",
|
131 |
+
"Greater",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
"Transpose",
|
133 |
+
"Constant",
|
134 |
+
"Concat"
|
135 |
],
|
136 |
"weight_type": "QInt8"
|
137 |
}
|