bert-base-uncased-squadv1-72.9-sparse
/
linear_layer_sparse_stats_total_23M_72.9_relative_sparsity.csv
,linear_id,shape,param_count,nnz_count | |
0,bert.encoder.layer.0.attention.self.query,"[768, 768]",589824,16252 | |
1,bert.encoder.layer.0.attention.self.key,"[768, 768]",589824,13054 | |
2,bert.encoder.layer.0.attention.self.value,"[768, 768]",589824,31394 | |
3,bert.encoder.layer.0.attention.output.dense,"[768, 768]",589824,44074 | |
4,bert.encoder.layer.0.intermediate.dense,"[3072, 768]",2359296,1159424 | |
5,bert.encoder.layer.0.output.dense,"[768, 3072]",2359296,1061306 | |
6,bert.encoder.layer.1.attention.self.query,"[768, 768]",589824,58240 | |
7,bert.encoder.layer.1.attention.self.key,"[768, 768]",589824,54729 | |
8,bert.encoder.layer.1.attention.self.value,"[768, 768]",589824,57604 | |
9,bert.encoder.layer.1.attention.output.dense,"[768, 768]",589824,65732 | |
10,bert.encoder.layer.1.intermediate.dense,"[3072, 768]",2359296,1230849 | |
11,bert.encoder.layer.1.output.dense,"[768, 3072]",2359296,1063370 | |
12,bert.encoder.layer.2.attention.self.query,"[768, 768]",589824,76074 | |
13,bert.encoder.layer.2.attention.self.key,"[768, 768]",589824,72798 | |
14,bert.encoder.layer.2.attention.self.value,"[768, 768]",589824,69367 | |
15,bert.encoder.layer.2.attention.output.dense,"[768, 768]",589824,68078 | |
16,bert.encoder.layer.2.intermediate.dense,"[3072, 768]",2359296,1260119 | |
17,bert.encoder.layer.2.output.dense,"[768, 3072]",2359296,1081396 | |
18,bert.encoder.layer.3.attention.self.query,"[768, 768]",589824,79954 | |
19,bert.encoder.layer.3.attention.self.key,"[768, 768]",589824,84045 | |
20,bert.encoder.layer.3.attention.self.value,"[768, 768]",589824,97323 | |
21,bert.encoder.layer.3.attention.output.dense,"[768, 768]",589824,95918 | |
22,bert.encoder.layer.3.intermediate.dense,"[3072, 768]",2359296,1263350 | |
23,bert.encoder.layer.3.output.dense,"[768, 3072]",2359296,1069440 | |
24,bert.encoder.layer.4.attention.self.query,"[768, 768]",589824,77306 | |
25,bert.encoder.layer.4.attention.self.key,"[768, 768]",589824,77738 | |
26,bert.encoder.layer.4.attention.self.value,"[768, 768]",589824,103253 | |
27,bert.encoder.layer.4.attention.output.dense,"[768, 768]",589824,103673 | |
28,bert.encoder.layer.4.intermediate.dense,"[3072, 768]",2359296,1253106 | |
29,bert.encoder.layer.4.output.dense,"[768, 3072]",2359296,1031342 | |
30,bert.encoder.layer.5.attention.self.query,"[768, 768]",589824,59269 | |
31,bert.encoder.layer.5.attention.self.key,"[768, 768]",589824,66896 | |
32,bert.encoder.layer.5.attention.self.value,"[768, 768]",589824,97790 | |
33,bert.encoder.layer.5.attention.output.dense,"[768, 768]",589824,93833 | |
34,bert.encoder.layer.5.intermediate.dense,"[3072, 768]",2359296,1263747 | |
35,bert.encoder.layer.5.output.dense,"[768, 3072]",2359296,1014643 | |
36,bert.encoder.layer.6.attention.self.query,"[768, 768]",589824,59641 | |
37,bert.encoder.layer.6.attention.self.key,"[768, 768]",589824,67416 | |
38,bert.encoder.layer.6.attention.self.value,"[768, 768]",589824,95731 | |
39,bert.encoder.layer.6.attention.output.dense,"[768, 768]",589824,85998 | |
40,bert.encoder.layer.6.intermediate.dense,"[3072, 768]",2359296,1181526 | |
41,bert.encoder.layer.6.output.dense,"[768, 3072]",2359296,920703 | |
42,bert.encoder.layer.7.attention.self.query,"[768, 768]",589824,44935 | |
43,bert.encoder.layer.7.attention.self.key,"[768, 768]",589824,56868 | |
44,bert.encoder.layer.7.attention.self.value,"[768, 768]",589824,89913 | |
45,bert.encoder.layer.7.attention.output.dense,"[768, 768]",589824,74029 | |
46,bert.encoder.layer.7.intermediate.dense,"[3072, 768]",2359296,995977 | |
47,bert.encoder.layer.7.output.dense,"[768, 3072]",2359296,777939 | |
48,bert.encoder.layer.8.attention.self.query,"[768, 768]",589824,49884 | |
49,bert.encoder.layer.8.attention.self.key,"[768, 768]",589824,57649 | |
50,bert.encoder.layer.8.attention.self.value,"[768, 768]",589824,92159 | |
51,bert.encoder.layer.8.attention.output.dense,"[768, 768]",589824,71883 | |
52,bert.encoder.layer.8.intermediate.dense,"[3072, 768]",2359296,757196 | |
53,bert.encoder.layer.8.output.dense,"[768, 3072]",2359296,568698 | |
54,bert.encoder.layer.9.attention.self.query,"[768, 768]",589824,68469 | |
55,bert.encoder.layer.9.attention.self.key,"[768, 768]",589824,69557 | |
56,bert.encoder.layer.9.attention.self.value,"[768, 768]",589824,32458 | |
57,bert.encoder.layer.9.attention.output.dense,"[768, 768]",589824,23895 | |
58,bert.encoder.layer.9.intermediate.dense,"[3072, 768]",2359296,359909 | |
59,bert.encoder.layer.9.output.dense,"[768, 3072]",2359296,245729 | |
60,bert.encoder.layer.10.attention.self.query,"[768, 768]",589824,42730 | |
61,bert.encoder.layer.10.attention.self.key,"[768, 768]",589824,44139 | |
62,bert.encoder.layer.10.attention.self.value,"[768, 768]",589824,17533 | |
63,bert.encoder.layer.10.attention.output.dense,"[768, 768]",589824,12605 | |
64,bert.encoder.layer.10.intermediate.dense,"[3072, 768]",2359296,233883 | |
65,bert.encoder.layer.10.output.dense,"[768, 3072]",2359296,130882 | |
66,bert.encoder.layer.11.attention.self.query,"[768, 768]",589824,11427 | |
67,bert.encoder.layer.11.attention.self.key,"[768, 768]",589824,14775 | |
68,bert.encoder.layer.11.attention.self.value,"[768, 768]",589824,6865 | |
69,bert.encoder.layer.11.attention.output.dense,"[768, 768]",589824,3223 | |
70,bert.encoder.layer.11.intermediate.dense,"[3072, 768]",2359296,190784 | |
71,bert.encoder.layer.11.output.dense,"[768, 3072]",2359296,71760 | |