Upload folder using huggingface_hub
Browse files- config.json +1 -0
- generation_config.json +1 -1
- 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
- quantize_config.json +64 -64
config.json
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
{
|
|
|
2 |
"architectures": [
|
3 |
"LlamaForCausalLM"
|
4 |
],
|
|
|
1 |
{
|
2 |
+
"_name_or_path": "Xenova/llama2.c-stories42M",
|
3 |
"architectures": [
|
4 |
"LlamaForCausalLM"
|
5 |
],
|
generation_config.json
CHANGED
@@ -2,5 +2,5 @@
|
|
2 |
"_from_model_config": true,
|
3 |
"bos_token_id": 1,
|
4 |
"eos_token_id": 2,
|
5 |
-
"transformers_version": "4.
|
6 |
}
|
|
|
2 |
"_from_model_config": true,
|
3 |
"bos_token_id": 1,
|
4 |
"eos_token_id": 2,
|
5 |
+
"transformers_version": "4.34.0.dev0"
|
6 |
}
|
onnx/decoder_model.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 233102386
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:57fecedae457d1811cd12abe22589f313aa15c1590e1e09bae3408052e6a74c7
|
3 |
size 233102386
|
onnx/decoder_model_merged.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 233913496
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:46464ab95c2c94ad1e493d0ca7746763f367343fce80cdcf7d0036911947564c
|
3 |
size 233913496
|
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 60022789
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:afb16b73d5c4f6e825abedc3ce65298ecd567e29c75dc98cc7f7b8dbf9e00074
|
3 |
size 60022789
|
onnx/decoder_model_quantized.onnx
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
size 59060350
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ea09156167b318bfb6149e325a18c79e95e0948175dd3fb496da53ca0589f070
|
3 |
size 59060350
|
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 233104923
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:47d1d207d61288f734b55448c70e5b344262c01087900f6c258b7805e0ed86a9
|
3 |
size 233104923
|
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 59062863
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ca6889c5883f8686fd4ba12ee26e025b490d7ffcbe1408c8b92f766e3df6f199
|
3 |
size 59062863
|
quantize_config.json
CHANGED
@@ -4,102 +4,102 @@
|
|
4 |
"per_model_config": {
|
5 |
"decoder_model": {
|
6 |
"op_types": [
|
7 |
-
"
|
8 |
-
"Concat",
|
9 |
-
"Cast",
|
10 |
-
"Identity",
|
11 |
-
"Div",
|
12 |
-
"Shape",
|
13 |
"Softmax",
|
14 |
-
"
|
15 |
-
"
|
|
|
16 |
"If",
|
17 |
-
"
|
|
|
|
|
18 |
"Mul",
|
19 |
-
"
|
|
|
20 |
"Slice",
|
21 |
-
"
|
|
|
|
|
22 |
"Range",
|
23 |
-
"
|
24 |
-
"
|
25 |
-
"
|
|
|
|
|
26 |
"Sigmoid",
|
27 |
"Where",
|
28 |
-
"Gather",
|
29 |
-
"Neg",
|
30 |
"Squeeze",
|
31 |
-
"
|
32 |
-
"
|
33 |
-
"
|
34 |
-
"
|
35 |
-
"Transpose"
|
36 |
],
|
37 |
"weight_type": "QInt8"
|
38 |
},
|
39 |
-
"
|
40 |
"op_types": [
|
41 |
-
"
|
42 |
-
"Concat",
|
43 |
-
"Cast",
|
44 |
-
"Identity",
|
45 |
-
"Div",
|
46 |
-
"Shape",
|
47 |
"Softmax",
|
48 |
-
"
|
49 |
-
"
|
|
|
50 |
"If",
|
51 |
-
"
|
|
|
|
|
52 |
"Mul",
|
53 |
-
"
|
|
|
54 |
"Slice",
|
55 |
-
"
|
|
|
56 |
"Range",
|
57 |
-
"
|
58 |
-
"
|
59 |
-
"
|
|
|
|
|
60 |
"Sigmoid",
|
61 |
"Where",
|
62 |
-
"Gather",
|
63 |
-
"Neg",
|
64 |
"Squeeze",
|
65 |
-
"
|
66 |
-
"
|
67 |
-
"
|
68 |
-
"
|
69 |
-
"Transpose"
|
70 |
],
|
71 |
"weight_type": "QInt8"
|
72 |
},
|
73 |
-
"
|
74 |
"op_types": [
|
75 |
-
"
|
76 |
-
"Concat",
|
77 |
-
"Cast",
|
78 |
-
"Identity",
|
79 |
-
"Div",
|
80 |
-
"Shape",
|
81 |
"Softmax",
|
82 |
-
"
|
83 |
-
"
|
|
|
84 |
"If",
|
85 |
-
"
|
|
|
|
|
86 |
"Mul",
|
|
|
|
|
87 |
"Slice",
|
88 |
-
"
|
|
|
|
|
89 |
"Range",
|
90 |
-
"
|
91 |
-
"
|
92 |
-
"
|
|
|
|
|
93 |
"Sigmoid",
|
94 |
"Where",
|
95 |
-
"Gather",
|
96 |
-
"Neg",
|
97 |
"Squeeze",
|
98 |
-
"
|
99 |
-
"
|
100 |
-
"
|
101 |
-
"
|
102 |
-
"Transpose"
|
103 |
],
|
104 |
"weight_type": "QInt8"
|
105 |
}
|
|
|
4 |
"per_model_config": {
|
5 |
"decoder_model": {
|
6 |
"op_types": [
|
7 |
+
"Sub",
|
|
|
|
|
|
|
|
|
|
|
8 |
"Softmax",
|
9 |
+
"Neg",
|
10 |
+
"Sqrt",
|
11 |
+
"Reshape",
|
12 |
"If",
|
13 |
+
"Shape",
|
14 |
+
"Transpose",
|
15 |
+
"Pow",
|
16 |
"Mul",
|
17 |
+
"MatMul",
|
18 |
+
"Gather",
|
19 |
"Slice",
|
20 |
+
"Less",
|
21 |
+
"ReduceMean",
|
22 |
+
"Concat",
|
23 |
"Range",
|
24 |
+
"Unsqueeze",
|
25 |
+
"Equal",
|
26 |
+
"Cast",
|
27 |
+
"ConstantOfShape",
|
28 |
+
"Add",
|
29 |
"Sigmoid",
|
30 |
"Where",
|
|
|
|
|
31 |
"Squeeze",
|
32 |
+
"Constant",
|
33 |
+
"Identity",
|
34 |
+
"Div",
|
35 |
+
"Expand"
|
|
|
36 |
],
|
37 |
"weight_type": "QInt8"
|
38 |
},
|
39 |
+
"decoder_with_past_model": {
|
40 |
"op_types": [
|
41 |
+
"Sub",
|
|
|
|
|
|
|
|
|
|
|
42 |
"Softmax",
|
43 |
+
"Neg",
|
44 |
+
"Sqrt",
|
45 |
+
"Reshape",
|
46 |
"If",
|
47 |
+
"Shape",
|
48 |
+
"Transpose",
|
49 |
+
"Pow",
|
50 |
"Mul",
|
51 |
+
"MatMul",
|
52 |
+
"Gather",
|
53 |
"Slice",
|
54 |
+
"ReduceMean",
|
55 |
+
"Concat",
|
56 |
"Range",
|
57 |
+
"Unsqueeze",
|
58 |
+
"Equal",
|
59 |
+
"Cast",
|
60 |
+
"ConstantOfShape",
|
61 |
+
"Add",
|
62 |
"Sigmoid",
|
63 |
"Where",
|
|
|
|
|
64 |
"Squeeze",
|
65 |
+
"Constant",
|
66 |
+
"Identity",
|
67 |
+
"Div",
|
68 |
+
"Expand"
|
|
|
69 |
],
|
70 |
"weight_type": "QInt8"
|
71 |
},
|
72 |
+
"decoder_model_merged": {
|
73 |
"op_types": [
|
74 |
+
"Sub",
|
|
|
|
|
|
|
|
|
|
|
75 |
"Softmax",
|
76 |
+
"Neg",
|
77 |
+
"Sqrt",
|
78 |
+
"Reshape",
|
79 |
"If",
|
80 |
+
"Shape",
|
81 |
+
"Transpose",
|
82 |
+
"Pow",
|
83 |
"Mul",
|
84 |
+
"MatMul",
|
85 |
+
"Gather",
|
86 |
"Slice",
|
87 |
+
"Less",
|
88 |
+
"ReduceMean",
|
89 |
+
"Concat",
|
90 |
"Range",
|
91 |
+
"Unsqueeze",
|
92 |
+
"Equal",
|
93 |
+
"Cast",
|
94 |
+
"ConstantOfShape",
|
95 |
+
"Add",
|
96 |
"Sigmoid",
|
97 |
"Where",
|
|
|
|
|
98 |
"Squeeze",
|
99 |
+
"Constant",
|
100 |
+
"Identity",
|
101 |
+
"Div",
|
102 |
+
"Expand"
|
|
|
103 |
],
|
104 |
"weight_type": "QInt8"
|
105 |
}
|