dkurt commited on
Commit
94fcbb4
1 Parent(s): 745f8cd

Add initial quantized model

Browse files
Files changed (6) hide show
  1. .gitattributes +2 -0
  2. README.md +5 -3
  3. config.json +117 -0
  4. nncf_config.json +36 -0
  5. ov_model.bin +3 -0
  6. ov_model.xml +3 -0
.gitattributes CHANGED
@@ -25,3 +25,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
25
  *.zip filter=lfs diff=lfs merge=lfs -text
26
  *.zstandard filter=lfs diff=lfs merge=lfs -text
27
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
25
  *.zip filter=lfs diff=lfs merge=lfs -text
26
  *.zstandard filter=lfs diff=lfs merge=lfs -text
27
  *tfevents* filter=lfs diff=lfs merge=lfs -text
28
+ ov_model.xml filter=lfs diff=lfs merge=lfs -text
29
+ ov_model.bin filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,3 +1,5 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
1
+ [anton-l/wav2vec2-base-ft-keyword-spotting](https://huggingface.co/anton-l/wav2vec2-base-ft-keyword-spotting) model quantized with [Optimum OpenVINO](https://github.com/dkurt/optimum-openvino/).
2
+
3
+ | Accuracy on eval (baseline) | Accuracy on eval (quantized) |
4
+ |-----------------------------|----------------------------------------|
5
+ | 0.9828 | 0.9553 (-0.0274) |
config.json ADDED
@@ -0,0 +1,117 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "facebook/wav2vec2-base",
3
+ "activation_dropout": 0.0,
4
+ "apply_spec_augment": true,
5
+ "architectures": [
6
+ "Wav2Vec2ForSequenceClassification"
7
+ ],
8
+ "attention_dropout": 0.1,
9
+ "bos_token_id": 1,
10
+ "classifier_proj_size": 256,
11
+ "codevector_dim": 256,
12
+ "contrastive_logits_temperature": 0.1,
13
+ "conv_bias": false,
14
+ "conv_dim": [
15
+ 512,
16
+ 512,
17
+ 512,
18
+ 512,
19
+ 512,
20
+ 512,
21
+ 512
22
+ ],
23
+ "conv_kernel": [
24
+ 10,
25
+ 3,
26
+ 3,
27
+ 3,
28
+ 3,
29
+ 2,
30
+ 2
31
+ ],
32
+ "conv_stride": [
33
+ 5,
34
+ 2,
35
+ 2,
36
+ 2,
37
+ 2,
38
+ 2,
39
+ 2
40
+ ],
41
+ "ctc_loss_reduction": "sum",
42
+ "ctc_zero_infinity": false,
43
+ "diversity_loss_weight": 0.1,
44
+ "do_stable_layer_norm": false,
45
+ "eos_token_id": 2,
46
+ "feat_extract_activation": "gelu",
47
+ "feat_extract_norm": "group",
48
+ "feat_proj_dropout": 0.1,
49
+ "feat_quantizer_dropout": 0.0,
50
+ "final_dropout": 0.0,
51
+ "finetuning_task": "audio-classification",
52
+ "freeze_feat_extract_train": true,
53
+ "hidden_act": "gelu",
54
+ "hidden_dropout": 0.1,
55
+ "hidden_size": 768,
56
+ "id2label": {
57
+ "0": "yes",
58
+ "1": "no",
59
+ "10": "_silence_",
60
+ "11": "_unknown_",
61
+ "2": "up",
62
+ "3": "down",
63
+ "4": "left",
64
+ "5": "right",
65
+ "6": "on",
66
+ "7": "off",
67
+ "8": "stop",
68
+ "9": "go"
69
+ },
70
+ "initializer_range": 0.02,
71
+ "intermediate_size": 3072,
72
+ "label2id": {
73
+ "_silence_": "10",
74
+ "_unknown_": "11",
75
+ "down": "3",
76
+ "go": "9",
77
+ "left": "4",
78
+ "no": "1",
79
+ "off": "7",
80
+ "on": "6",
81
+ "right": "5",
82
+ "stop": "8",
83
+ "up": "2",
84
+ "yes": "0"
85
+ },
86
+ "layer_norm_eps": 1e-05,
87
+ "layerdrop": 0.05,
88
+ "mask_channel_length": 10,
89
+ "mask_channel_min_space": 1,
90
+ "mask_channel_other": 0.0,
91
+ "mask_channel_prob": 0.0,
92
+ "mask_channel_selection": "static",
93
+ "mask_feature_length": 10,
94
+ "mask_feature_prob": 0.0,
95
+ "mask_time_length": 10,
96
+ "mask_time_min_space": 1,
97
+ "mask_time_other": 0.0,
98
+ "mask_time_prob": 0.05,
99
+ "mask_time_selection": "static",
100
+ "model_type": "wav2vec2",
101
+ "no_mask_channel_overlap": false,
102
+ "no_mask_time_overlap": false,
103
+ "num_attention_heads": 12,
104
+ "num_codevector_groups": 2,
105
+ "num_codevectors_per_group": 320,
106
+ "num_conv_pos_embedding_groups": 16,
107
+ "num_conv_pos_embeddings": 128,
108
+ "num_feat_extract_layers": 7,
109
+ "num_hidden_layers": 12,
110
+ "num_negatives": 100,
111
+ "pad_token_id": 0,
112
+ "proj_codevector_dim": 256,
113
+ "torch_dtype": "float32",
114
+ "transformers_version": "4.12.0.dev0",
115
+ "use_weighted_layer_sum": false,
116
+ "vocab_size": 32
117
+ }
nncf_config.json ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "input_info": [
3
+ {
4
+ "sample_size": [1, 16000],
5
+ "type": "float"
6
+ }
7
+ ],
8
+ "compression": {
9
+ "algorithm": "quantization",
10
+ "initializer": {
11
+ "range": {
12
+ "num_init_samples": 32,
13
+ "type": "percentile",
14
+ "params":
15
+ {
16
+ "min_percentile": 0.01,
17
+ "max_percentile": 99.99
18
+ }
19
+ },
20
+
21
+ "batchnorm_adaptation": {
22
+ "num_bn_adaptation_samples": 200
23
+ }
24
+ },
25
+ "activations":
26
+ {
27
+ "mode": "asymmetric"
28
+ },
29
+ "weights":
30
+ {
31
+ "mode": "symmetric",
32
+ "signed": true,
33
+ "per_channel": true
34
+ }
35
+ }
36
+ }
ov_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:229e84a458770e5bf4de7dcdcd5e0d55f8bac42287e5c8d2304c123eb08dcb90
3
+ size 95305472
ov_model.xml ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7006a197f10bac4b220e8aea721946acb7d2e2a6539d8d0b24c8397a24ae18cc
3
+ size 1218568