Upload folder using huggingface_hub
Browse files
onnx/decoder_model.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 sha256:9d3466d176ef856af51c6fc5d3ce88ae506a1cb21c4ef127a85580a902542306
|
3 |
+
size 118406656
|
onnx/decoder_model_merged.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 sha256:0589bdc5c4bce7f9dc444642f4735b8c4b862e5dc9a7abb9dad12fc70ee935c9
|
3 |
+
size 118579599
|
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
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d72f5d151c6418ca344c88f7ddb5e489d29b62b0c2a80dbb51990ae6822997ce
|
3 |
+
size 30727766
|
onnx/decoder_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 sha256:9c780e74cd3b127beb7f7a0c3ea5d1a688927b478a65348126e19b56c7d9c15b
|
3 |
+
size 30460102
|
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
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:d2e075fd460f17a6561f85149705f0bbf08c63b939182dd1abcdce1f54ae6fbb
|
3 |
+
size 113654551
|
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
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c6d764b0cbab6a9d34b06adcd1589f587c6d3b33c4312290e6c6eb84c0a44e95
|
3 |
+
size 29229120
|
quantize_config.json
CHANGED
@@ -2,112 +2,112 @@
|
|
2 |
"per_channel": false,
|
3 |
"reduce_range": false,
|
4 |
"per_model_config": {
|
5 |
-
"
|
6 |
"op_types": [
|
7 |
-
"
|
8 |
-
"
|
9 |
-
"Concat",
|
10 |
-
"Erf",
|
11 |
"Add",
|
|
|
|
|
12 |
"ConstantOfShape",
|
13 |
-
"Reshape",
|
14 |
-
"Transpose",
|
15 |
-
"Less",
|
16 |
-
"Equal",
|
17 |
-
"Gather",
|
18 |
-
"If",
|
19 |
-
"Where",
|
20 |
-
"Softmax",
|
21 |
-
"Div",
|
22 |
-
"Unsqueeze",
|
23 |
-
"Cast",
|
24 |
-
"Mul",
|
25 |
"Shape",
|
|
|
26 |
"Slice",
|
27 |
-
"
|
|
|
28 |
"Pow",
|
29 |
-
"
|
|
|
|
|
|
|
30 |
"Constant",
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
"MatMul",
|
32 |
"Range",
|
33 |
-
"
|
34 |
],
|
35 |
"weight_type": "QInt8"
|
36 |
},
|
37 |
-
"
|
38 |
"op_types": [
|
39 |
-
"
|
40 |
-
"
|
41 |
-
"Concat",
|
42 |
-
"Erf",
|
43 |
"Add",
|
|
|
|
|
44 |
"ConstantOfShape",
|
45 |
-
"Reshape",
|
46 |
-
"Transpose",
|
47 |
-
"Less",
|
48 |
-
"Equal",
|
49 |
-
"Gather",
|
50 |
-
"Where",
|
51 |
-
"Softmax",
|
52 |
-
"Div",
|
53 |
-
"Unsqueeze",
|
54 |
-
"Cast",
|
55 |
-
"Mul",
|
56 |
"Shape",
|
|
|
57 |
"Slice",
|
58 |
-
"
|
|
|
59 |
"Pow",
|
60 |
-
"
|
|
|
|
|
|
|
61 |
"Constant",
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
"MatMul",
|
|
|
63 |
"Range",
|
64 |
-
"
|
65 |
],
|
66 |
"weight_type": "QInt8"
|
67 |
},
|
68 |
"decoder_with_past_model": {
|
69 |
"op_types": [
|
70 |
-
"
|
|
|
|
|
71 |
"Shape",
|
72 |
-
"
|
73 |
"Add",
|
74 |
-
"
|
75 |
-
"
|
76 |
-
"ReduceMean",
|
77 |
"Constant",
|
78 |
-
"
|
|
|
|
|
79 |
"Reshape",
|
|
|
80 |
"MatMul",
|
81 |
-
"
|
82 |
-
"Softmax",
|
83 |
-
"Sub",
|
84 |
"Erf",
|
85 |
-
"Gather"
|
86 |
-
"Div",
|
87 |
-
"Unsqueeze"
|
88 |
],
|
89 |
"weight_type": "QInt8"
|
90 |
},
|
91 |
"encoder_model": {
|
92 |
"op_types": [
|
93 |
-
"Mul",
|
94 |
-
"Conv",
|
95 |
-
"Shape",
|
96 |
-
"Add",
|
97 |
"Sqrt",
|
98 |
-
"
|
99 |
-
"
|
100 |
-
"Constant",
|
101 |
-
"MatMul",
|
102 |
"Concat",
|
103 |
-
"
|
|
|
104 |
"Transpose",
|
105 |
-
"
|
|
|
106 |
"Sub",
|
107 |
-
"
|
108 |
"Gather",
|
109 |
-
"
|
110 |
-
"Unsqueeze"
|
|
|
|
|
|
|
|
|
111 |
],
|
112 |
"weight_type": "QUInt8"
|
113 |
}
|
|
|
2 |
"per_channel": false,
|
3 |
"reduce_range": false,
|
4 |
"per_model_config": {
|
5 |
+
"decoder_model": {
|
6 |
"op_types": [
|
7 |
+
"Sqrt",
|
8 |
+
"Equal",
|
|
|
|
|
9 |
"Add",
|
10 |
+
"Sub",
|
11 |
+
"ReduceMean",
|
12 |
"ConstantOfShape",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
"Shape",
|
14 |
+
"Div",
|
15 |
"Slice",
|
16 |
+
"Expand",
|
17 |
+
"Squeeze",
|
18 |
"Pow",
|
19 |
+
"Less",
|
20 |
+
"Softmax",
|
21 |
+
"Concat",
|
22 |
+
"Transpose",
|
23 |
"Constant",
|
24 |
+
"Unsqueeze",
|
25 |
+
"Mul",
|
26 |
+
"Erf",
|
27 |
+
"Gather",
|
28 |
+
"Where",
|
29 |
+
"Reshape",
|
30 |
"MatMul",
|
31 |
"Range",
|
32 |
+
"Cast"
|
33 |
],
|
34 |
"weight_type": "QInt8"
|
35 |
},
|
36 |
+
"decoder_model_merged": {
|
37 |
"op_types": [
|
38 |
+
"Sqrt",
|
39 |
+
"Equal",
|
|
|
|
|
40 |
"Add",
|
41 |
+
"Sub",
|
42 |
+
"ReduceMean",
|
43 |
"ConstantOfShape",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
"Shape",
|
45 |
+
"Div",
|
46 |
"Slice",
|
47 |
+
"Expand",
|
48 |
+
"Squeeze",
|
49 |
"Pow",
|
50 |
+
"Less",
|
51 |
+
"Softmax",
|
52 |
+
"Concat",
|
53 |
+
"Transpose",
|
54 |
"Constant",
|
55 |
+
"Unsqueeze",
|
56 |
+
"Mul",
|
57 |
+
"Erf",
|
58 |
+
"Gather",
|
59 |
+
"Where",
|
60 |
+
"Reshape",
|
61 |
"MatMul",
|
62 |
+
"If",
|
63 |
"Range",
|
64 |
+
"Cast"
|
65 |
],
|
66 |
"weight_type": "QInt8"
|
67 |
},
|
68 |
"decoder_with_past_model": {
|
69 |
"op_types": [
|
70 |
+
"Sqrt",
|
71 |
+
"Softmax",
|
72 |
+
"Concat",
|
73 |
"Shape",
|
74 |
+
"Div",
|
75 |
"Add",
|
76 |
+
"Slice",
|
77 |
+
"Transpose",
|
|
|
78 |
"Constant",
|
79 |
+
"Unsqueeze",
|
80 |
+
"Sub",
|
81 |
+
"Pow",
|
82 |
"Reshape",
|
83 |
+
"Mul",
|
84 |
"MatMul",
|
85 |
+
"ReduceMean",
|
|
|
|
|
86 |
"Erf",
|
87 |
+
"Gather"
|
|
|
|
|
88 |
],
|
89 |
"weight_type": "QInt8"
|
90 |
},
|
91 |
"encoder_model": {
|
92 |
"op_types": [
|
|
|
|
|
|
|
|
|
93 |
"Sqrt",
|
94 |
+
"Softmax",
|
95 |
+
"Conv",
|
|
|
|
|
96 |
"Concat",
|
97 |
+
"Div",
|
98 |
+
"Add",
|
99 |
"Transpose",
|
100 |
+
"Constant",
|
101 |
+
"Pow",
|
102 |
"Sub",
|
103 |
+
"Mul",
|
104 |
"Gather",
|
105 |
+
"MatMul",
|
106 |
+
"Unsqueeze",
|
107 |
+
"ReduceMean",
|
108 |
+
"Erf",
|
109 |
+
"Reshape",
|
110 |
+
"Shape"
|
111 |
],
|
112 |
"weight_type": "QUInt8"
|
113 |
}
|