Vui Seng Chua commited on
Commit
717f4ec
·
1 Parent(s): 95fef9e

add jpqd-lt-r0.02-8eph-optimum-cfg-QEmb

Browse files
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/mo-pruned-ir/nncf_network.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee01233efa3938858ccd58c64ecb6b4eb11e11c7aad5299629be056d6fc89b40
3
+ size 74602984
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/mo-pruned-ir/nncf_network.mapping ADDED
The diff for this file is too large to render. See raw diff
 
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/mo-pruned-ir/nncf_network.xml ADDED
The diff for this file is too large to render. See raw diff
 
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/nncf_network.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:28d9a6769085df956591b45bd663c36623247ab4127d4150e78f946a613ed8fc
3
+ size 109593096
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/nncf_network.mapping ADDED
The diff for this file is too large to render. See raw diff
 
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/nncf_network.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a8e872e9e20836ddf9b54250a2e0cd701e16e2075d110c0a55ac4ac7cf968191
3
+ size 437092459
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/nncf_network.xml ADDED
The diff for this file is too large to render. See raw diff
 
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/openvino_config.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "compression": [
3
+ {
4
+ "algorithm": "movement_sparsity",
5
+ "ignored_scopes": [
6
+ "{re}.*NNCFEmbedding",
7
+ "{re}.*qa_outputs*",
8
+ "{re}.*LayerNorm.*"
9
+ ],
10
+ "params": {
11
+ "enable_structured_masking": true,
12
+ "importance_regularization_factor": 0.02,
13
+ "warmup_end_epoch": 4,
14
+ "warmup_start_epoch": 1
15
+ },
16
+ "sparse_structure_by_scopes": [
17
+ {
18
+ "mode": "block",
19
+ "sparse_factors": [
20
+ 32,
21
+ 32
22
+ ],
23
+ "target_scopes": "{re}.*BertAttention*"
24
+ },
25
+ {
26
+ "axis": 0,
27
+ "mode": "per_dim",
28
+ "target_scopes": "{re}.*BertIntermediate.*"
29
+ },
30
+ {
31
+ "axis": 1,
32
+ "mode": "per_dim",
33
+ "target_scopes": "{re}.*BertOutput.*"
34
+ }
35
+ ]
36
+ },
37
+ {
38
+ "algorithm": "quantization",
39
+ "export_to_onnx_standard_ops": false,
40
+ "ignored_scopes": [
41
+ "{re}.*__add___[0-1]",
42
+ "{re}.*layer_norm_0",
43
+ "{re}.*matmul_1",
44
+ "{re}.*__truediv__*"
45
+ ],
46
+ "initializer": {
47
+ "batchnorm_adaptation": {
48
+ "num_bn_adaptation_samples": 200
49
+ },
50
+ "range": {
51
+ "num_init_samples": 32,
52
+ "params": {
53
+ "max_percentile": 99.99,
54
+ "min_percentile": 0.01
55
+ },
56
+ "type": "percentile"
57
+ }
58
+ },
59
+ "overflow_fix": "disable",
60
+ "preset": "mixed",
61
+ "scope_overrides": {
62
+ "activations": {
63
+ "{re}.*matmul_0": {
64
+ "mode": "symmetric"
65
+ }
66
+ }
67
+ }
68
+ }
69
+ ],
70
+ "input_info": [
71
+ {
72
+ "keyword": "input_ids",
73
+ "sample_size": [
74
+ 16,
75
+ 384
76
+ ],
77
+ "type": "long"
78
+ },
79
+ {
80
+ "keyword": "token_type_ids",
81
+ "sample_size": [
82
+ 16,
83
+ 384
84
+ ],
85
+ "type": "long"
86
+ },
87
+ {
88
+ "keyword": "attention_mask",
89
+ "sample_size": [
90
+ 16,
91
+ 384
92
+ ],
93
+ "type": "long"
94
+ }
95
+ ],
96
+ "optimum_version": "1.5.2",
97
+ "save_onnx_model": false,
98
+ "transformers_version": "4.25.1"
99
+ }
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/openvino_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:99dbb80c0e253f3f64d4f14dd48846312b79063694d09bde2805a1d22ed32a39
3
+ size 109594088
ir/jpqd-lt-r0.02-8eph-optimum-cfg-QEmb/openvino_model.xml ADDED
The diff for this file is too large to render. See raw diff