Initial commit

This commit is contained in:
Breadway 2026-07-02 20:59:18 +08:00
commit 2778f14574
29 changed files with 7110 additions and 0 deletions

151
quantized_info.csv Normal file
View file

@ -0,0 +1,151 @@
/home/breadway/.cache/breadsearch/models/model.onnx quantization quantization info,2026-06-24 18:14:20
quantization stage,time consumed(s),sub stage,time consumed(s)
pre process,5.43
,,calibration: collect data (onnx inference + numpy statistics),16.60
,,calibration: compute data,0.01
calibration (collect data + compute data),19.75
static quantization,0.81
post process(including finetuning),0.00
e2e,40.621040006999465
Node Name,Op Type,Activation,Weights,Bias
/encoder/layers.0/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.0/attn/Reshape,Reshape,INT8,,
/encoder/layers.0/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.0/Add,Add,INT8,,
/encoder/layers.0/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.0/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.0/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.0/mlp/Mul,Mul,INT8,,
/encoder/layers.0/mlp/Mul_1,Mul,INT8,,
/encoder/layers.0/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.0/Add_1,Add,INT8,,
/encoder/layers.1/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.1/attn/Reshape,Reshape,INT8,,
/encoder/layers.1/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.1/Add,Add,INT8,,
/encoder/layers.1/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.1/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.1/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.1/mlp/Mul,Mul,INT8,,
/encoder/layers.1/mlp/Mul_1,Mul,INT8,,
/encoder/layers.1/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.1/Add_1,Add,INT8,,
/encoder/layers.2/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.2/attn/Reshape,Reshape,INT8,,
/encoder/layers.2/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.2/Add,Add,INT8,,
/encoder/layers.2/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.2/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.2/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.2/mlp/Mul,Mul,INT8,,
/encoder/layers.2/mlp/Mul_1,Mul,INT8,,
/encoder/layers.2/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.2/Add_1,Add,INT8,,
/encoder/layers.3/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.3/attn/Reshape,Reshape,INT8,,
/encoder/layers.3/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.3/Add,Add,INT8,,
/encoder/layers.3/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.3/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.3/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.3/mlp/Mul,Mul,INT8,,
/encoder/layers.3/mlp/Mul_1,Mul,INT8,,
/encoder/layers.3/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.3/Add_1,Add,INT8,,
/encoder/layers.4/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.4/attn/Reshape,Reshape,INT8,,
/encoder/layers.4/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.4/Add,Add,INT8,,
/encoder/layers.4/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.4/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.4/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.4/mlp/Mul,Mul,INT8,,
/encoder/layers.4/mlp/Mul_1,Mul,INT8,,
/encoder/layers.4/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.4/Add_1,Add,INT8,,
/encoder/layers.5/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.5/attn/Reshape,Reshape,INT8,,
/encoder/layers.5/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.5/Add,Add,INT8,,
/encoder/layers.5/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.5/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.5/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.5/mlp/Mul,Mul,INT8,,
/encoder/layers.5/mlp/Mul_1,Mul,INT8,,
/encoder/layers.5/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.5/Add_1,Add,INT8,,
/encoder/layers.6/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.6/attn/Reshape,Reshape,INT8,,
/encoder/layers.6/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.6/Add,Add,INT8,,
/encoder/layers.6/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.6/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.6/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.6/mlp/Mul,Mul,INT8,,
/encoder/layers.6/mlp/Mul_1,Mul,INT8,,
/encoder/layers.6/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.6/Add_1,Add,INT8,,
/encoder/layers.7/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.7/attn/Reshape,Reshape,INT8,,
/encoder/layers.7/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.7/Add,Add,INT8,,
/encoder/layers.7/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.7/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.7/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.7/mlp/Mul,Mul,INT8,,
/encoder/layers.7/mlp/Mul_1,Mul,INT8,,
/encoder/layers.7/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.7/Add_1,Add,INT8,,
/encoder/layers.8/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.8/attn/Reshape,Reshape,INT8,,
/encoder/layers.8/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.8/Add,Add,INT8,,
/encoder/layers.8/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.8/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.8/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.8/mlp/Mul,Mul,INT8,,
/encoder/layers.8/mlp/Mul_1,Mul,INT8,,
/encoder/layers.8/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.8/Add_1,Add,INT8,,
/encoder/layers.9/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.9/attn/Reshape,Reshape,INT8,,
/encoder/layers.9/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.9/Add,Add,INT8,,
/encoder/layers.9/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.9/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.9/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.9/mlp/Mul,Mul,INT8,,
/encoder/layers.9/mlp/Mul_1,Mul,INT8,,
/encoder/layers.9/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.9/Add_1,Add,INT8,,
/encoder/layers.10/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.10/attn/Reshape,Reshape,INT8,,
/encoder/layers.10/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.10/Add,Add,INT8,,
/encoder/layers.10/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.10/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.10/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.10/mlp/Mul,Mul,INT8,,
/encoder/layers.10/mlp/Mul_1,Mul,INT8,,
/encoder/layers.10/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.10/Add_1,Add,INT8,,
/encoder/layers.11/attn/Wqkv/MatMul,MatMul,INT8,INT8,
/encoder/layers.11/attn/Reshape,Reshape,INT8,,
/encoder/layers.11/attn/out_proj/MatMul,MatMul,INT8,INT8,
/encoder/layers.11/Add,Add,INT8,,
/encoder/layers.11/mlp/fc12/MatMul,MatMul,INT8,INT8,
/encoder/layers.11/mlp/fc11/MatMul,MatMul,INT8,INT8,
/encoder/layers.11/mlp/Sigmoid,Sigmoid,INT8,,
/encoder/layers.11/mlp/Mul,Mul,INT8,,
/encoder/layers.11/mlp/Mul_1,Mul,INT8,,
/encoder/layers.11/mlp/fc2/MatMul,MatMul,INT8,INT8,
/encoder/layers.11/Add_1,Add,INT8,,
Op Type,Activation,Weights,Bias
MatMul,INT8(60),INT8(60),
Reshape,INT8(12),,
Add,INT8(24),,
Sigmoid,INT8(12),,
Mul,INT8(24),,
1 /home/breadway/.cache/breadsearch/models/model.onnx quantization quantization info,2026-06-24 18:14:20
2 quantization stage,time consumed(s),sub stage,time consumed(s)
3 pre process,5.43
4 ,,calibration: collect data (onnx inference + numpy statistics),16.60
5 ,,calibration: compute data,0.01
6 calibration (collect data + compute data),19.75
7 static quantization,0.81
8 post process(including finetuning),0.00
9 e2e,40.621040006999465
10 Node Name,Op Type,Activation,Weights,Bias
11 /encoder/layers.0/attn/Wqkv/MatMul,MatMul,INT8,INT8,
12 /encoder/layers.0/attn/Reshape,Reshape,INT8,,
13 /encoder/layers.0/attn/out_proj/MatMul,MatMul,INT8,INT8,
14 /encoder/layers.0/Add,Add,INT8,,
15 /encoder/layers.0/mlp/fc12/MatMul,MatMul,INT8,INT8,
16 /encoder/layers.0/mlp/fc11/MatMul,MatMul,INT8,INT8,
17 /encoder/layers.0/mlp/Sigmoid,Sigmoid,INT8,,
18 /encoder/layers.0/mlp/Mul,Mul,INT8,,
19 /encoder/layers.0/mlp/Mul_1,Mul,INT8,,
20 /encoder/layers.0/mlp/fc2/MatMul,MatMul,INT8,INT8,
21 /encoder/layers.0/Add_1,Add,INT8,,
22 /encoder/layers.1/attn/Wqkv/MatMul,MatMul,INT8,INT8,
23 /encoder/layers.1/attn/Reshape,Reshape,INT8,,
24 /encoder/layers.1/attn/out_proj/MatMul,MatMul,INT8,INT8,
25 /encoder/layers.1/Add,Add,INT8,,
26 /encoder/layers.1/mlp/fc12/MatMul,MatMul,INT8,INT8,
27 /encoder/layers.1/mlp/fc11/MatMul,MatMul,INT8,INT8,
28 /encoder/layers.1/mlp/Sigmoid,Sigmoid,INT8,,
29 /encoder/layers.1/mlp/Mul,Mul,INT8,,
30 /encoder/layers.1/mlp/Mul_1,Mul,INT8,,
31 /encoder/layers.1/mlp/fc2/MatMul,MatMul,INT8,INT8,
32 /encoder/layers.1/Add_1,Add,INT8,,
33 /encoder/layers.2/attn/Wqkv/MatMul,MatMul,INT8,INT8,
34 /encoder/layers.2/attn/Reshape,Reshape,INT8,,
35 /encoder/layers.2/attn/out_proj/MatMul,MatMul,INT8,INT8,
36 /encoder/layers.2/Add,Add,INT8,,
37 /encoder/layers.2/mlp/fc12/MatMul,MatMul,INT8,INT8,
38 /encoder/layers.2/mlp/fc11/MatMul,MatMul,INT8,INT8,
39 /encoder/layers.2/mlp/Sigmoid,Sigmoid,INT8,,
40 /encoder/layers.2/mlp/Mul,Mul,INT8,,
41 /encoder/layers.2/mlp/Mul_1,Mul,INT8,,
42 /encoder/layers.2/mlp/fc2/MatMul,MatMul,INT8,INT8,
43 /encoder/layers.2/Add_1,Add,INT8,,
44 /encoder/layers.3/attn/Wqkv/MatMul,MatMul,INT8,INT8,
45 /encoder/layers.3/attn/Reshape,Reshape,INT8,,
46 /encoder/layers.3/attn/out_proj/MatMul,MatMul,INT8,INT8,
47 /encoder/layers.3/Add,Add,INT8,,
48 /encoder/layers.3/mlp/fc12/MatMul,MatMul,INT8,INT8,
49 /encoder/layers.3/mlp/fc11/MatMul,MatMul,INT8,INT8,
50 /encoder/layers.3/mlp/Sigmoid,Sigmoid,INT8,,
51 /encoder/layers.3/mlp/Mul,Mul,INT8,,
52 /encoder/layers.3/mlp/Mul_1,Mul,INT8,,
53 /encoder/layers.3/mlp/fc2/MatMul,MatMul,INT8,INT8,
54 /encoder/layers.3/Add_1,Add,INT8,,
55 /encoder/layers.4/attn/Wqkv/MatMul,MatMul,INT8,INT8,
56 /encoder/layers.4/attn/Reshape,Reshape,INT8,,
57 /encoder/layers.4/attn/out_proj/MatMul,MatMul,INT8,INT8,
58 /encoder/layers.4/Add,Add,INT8,,
59 /encoder/layers.4/mlp/fc12/MatMul,MatMul,INT8,INT8,
60 /encoder/layers.4/mlp/fc11/MatMul,MatMul,INT8,INT8,
61 /encoder/layers.4/mlp/Sigmoid,Sigmoid,INT8,,
62 /encoder/layers.4/mlp/Mul,Mul,INT8,,
63 /encoder/layers.4/mlp/Mul_1,Mul,INT8,,
64 /encoder/layers.4/mlp/fc2/MatMul,MatMul,INT8,INT8,
65 /encoder/layers.4/Add_1,Add,INT8,,
66 /encoder/layers.5/attn/Wqkv/MatMul,MatMul,INT8,INT8,
67 /encoder/layers.5/attn/Reshape,Reshape,INT8,,
68 /encoder/layers.5/attn/out_proj/MatMul,MatMul,INT8,INT8,
69 /encoder/layers.5/Add,Add,INT8,,
70 /encoder/layers.5/mlp/fc12/MatMul,MatMul,INT8,INT8,
71 /encoder/layers.5/mlp/fc11/MatMul,MatMul,INT8,INT8,
72 /encoder/layers.5/mlp/Sigmoid,Sigmoid,INT8,,
73 /encoder/layers.5/mlp/Mul,Mul,INT8,,
74 /encoder/layers.5/mlp/Mul_1,Mul,INT8,,
75 /encoder/layers.5/mlp/fc2/MatMul,MatMul,INT8,INT8,
76 /encoder/layers.5/Add_1,Add,INT8,,
77 /encoder/layers.6/attn/Wqkv/MatMul,MatMul,INT8,INT8,
78 /encoder/layers.6/attn/Reshape,Reshape,INT8,,
79 /encoder/layers.6/attn/out_proj/MatMul,MatMul,INT8,INT8,
80 /encoder/layers.6/Add,Add,INT8,,
81 /encoder/layers.6/mlp/fc12/MatMul,MatMul,INT8,INT8,
82 /encoder/layers.6/mlp/fc11/MatMul,MatMul,INT8,INT8,
83 /encoder/layers.6/mlp/Sigmoid,Sigmoid,INT8,,
84 /encoder/layers.6/mlp/Mul,Mul,INT8,,
85 /encoder/layers.6/mlp/Mul_1,Mul,INT8,,
86 /encoder/layers.6/mlp/fc2/MatMul,MatMul,INT8,INT8,
87 /encoder/layers.6/Add_1,Add,INT8,,
88 /encoder/layers.7/attn/Wqkv/MatMul,MatMul,INT8,INT8,
89 /encoder/layers.7/attn/Reshape,Reshape,INT8,,
90 /encoder/layers.7/attn/out_proj/MatMul,MatMul,INT8,INT8,
91 /encoder/layers.7/Add,Add,INT8,,
92 /encoder/layers.7/mlp/fc12/MatMul,MatMul,INT8,INT8,
93 /encoder/layers.7/mlp/fc11/MatMul,MatMul,INT8,INT8,
94 /encoder/layers.7/mlp/Sigmoid,Sigmoid,INT8,,
95 /encoder/layers.7/mlp/Mul,Mul,INT8,,
96 /encoder/layers.7/mlp/Mul_1,Mul,INT8,,
97 /encoder/layers.7/mlp/fc2/MatMul,MatMul,INT8,INT8,
98 /encoder/layers.7/Add_1,Add,INT8,,
99 /encoder/layers.8/attn/Wqkv/MatMul,MatMul,INT8,INT8,
100 /encoder/layers.8/attn/Reshape,Reshape,INT8,,
101 /encoder/layers.8/attn/out_proj/MatMul,MatMul,INT8,INT8,
102 /encoder/layers.8/Add,Add,INT8,,
103 /encoder/layers.8/mlp/fc12/MatMul,MatMul,INT8,INT8,
104 /encoder/layers.8/mlp/fc11/MatMul,MatMul,INT8,INT8,
105 /encoder/layers.8/mlp/Sigmoid,Sigmoid,INT8,,
106 /encoder/layers.8/mlp/Mul,Mul,INT8,,
107 /encoder/layers.8/mlp/Mul_1,Mul,INT8,,
108 /encoder/layers.8/mlp/fc2/MatMul,MatMul,INT8,INT8,
109 /encoder/layers.8/Add_1,Add,INT8,,
110 /encoder/layers.9/attn/Wqkv/MatMul,MatMul,INT8,INT8,
111 /encoder/layers.9/attn/Reshape,Reshape,INT8,,
112 /encoder/layers.9/attn/out_proj/MatMul,MatMul,INT8,INT8,
113 /encoder/layers.9/Add,Add,INT8,,
114 /encoder/layers.9/mlp/fc12/MatMul,MatMul,INT8,INT8,
115 /encoder/layers.9/mlp/fc11/MatMul,MatMul,INT8,INT8,
116 /encoder/layers.9/mlp/Sigmoid,Sigmoid,INT8,,
117 /encoder/layers.9/mlp/Mul,Mul,INT8,,
118 /encoder/layers.9/mlp/Mul_1,Mul,INT8,,
119 /encoder/layers.9/mlp/fc2/MatMul,MatMul,INT8,INT8,
120 /encoder/layers.9/Add_1,Add,INT8,,
121 /encoder/layers.10/attn/Wqkv/MatMul,MatMul,INT8,INT8,
122 /encoder/layers.10/attn/Reshape,Reshape,INT8,,
123 /encoder/layers.10/attn/out_proj/MatMul,MatMul,INT8,INT8,
124 /encoder/layers.10/Add,Add,INT8,,
125 /encoder/layers.10/mlp/fc12/MatMul,MatMul,INT8,INT8,
126 /encoder/layers.10/mlp/fc11/MatMul,MatMul,INT8,INT8,
127 /encoder/layers.10/mlp/Sigmoid,Sigmoid,INT8,,
128 /encoder/layers.10/mlp/Mul,Mul,INT8,,
129 /encoder/layers.10/mlp/Mul_1,Mul,INT8,,
130 /encoder/layers.10/mlp/fc2/MatMul,MatMul,INT8,INT8,
131 /encoder/layers.10/Add_1,Add,INT8,,
132 /encoder/layers.11/attn/Wqkv/MatMul,MatMul,INT8,INT8,
133 /encoder/layers.11/attn/Reshape,Reshape,INT8,,
134 /encoder/layers.11/attn/out_proj/MatMul,MatMul,INT8,INT8,
135 /encoder/layers.11/Add,Add,INT8,,
136 /encoder/layers.11/mlp/fc12/MatMul,MatMul,INT8,INT8,
137 /encoder/layers.11/mlp/fc11/MatMul,MatMul,INT8,INT8,
138 /encoder/layers.11/mlp/Sigmoid,Sigmoid,INT8,,
139 /encoder/layers.11/mlp/Mul,Mul,INT8,,
140 /encoder/layers.11/mlp/Mul_1,Mul,INT8,,
141 /encoder/layers.11/mlp/fc2/MatMul,MatMul,INT8,INT8,
142 /encoder/layers.11/Add_1,Add,INT8,,
143 Op Type,Activation,Weights,Bias
144 MatMul,INT8(60),INT8(60),
145 Reshape,INT8(12),,
146 Add,INT8(24),,
147 Sigmoid,INT8(12),,
148 Mul,INT8(24),,