Hugging Face's logo

ZenMan67
/
asr_checkpoint_v1

Model card Files
xet
 Community
asr_checkpoint_v1
File size: 12,509 Bytes
4380993
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
2025-10-16 05:54:40,737 - train - INFO - ConformerModel(
  (subsample): ConvSubsampling(
    (conv): Sequential(
      (0): Conv2d(1, 256, kernel_size=(3, 3), stride=(1, 2), padding=(1, 1))
      (1): ReLU()
      (2): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 2), padding=(1, 1))
      (3): ReLU()
    )
  )
  (pre_proj): LinearProjection(
    (proj): Linear(in_features=65536, out_features=512, bias=True)
  )
  (pos_enc): RelativePositionalEncoding()
  (blocks): ModuleList(
    (0-15): 16 x ConformerBlock(
      (ff1): FeedForwardModule(
        (layer_norm): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
        (linear1): Linear(in_features=512, out_features=2048, bias=True)
        (swish): Swish()
        (dropout): Dropout(p=0.1, inplace=False)
        (linear2): Linear(in_features=2048, out_features=512, bias=True)
        (dropout2): Dropout(p=0.1, inplace=False)
      )
      (mhsa): MultiHeadSelfAttention(
        (layer_norm): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
        (attn): MultiheadAttention(
          (out_proj): NonDynamicallyQuantizableLinear(in_features=512, out_features=512, bias=True)
        )
        (dropout): Dropout(p=0.1, inplace=False)
      )
      (conv): ConformerConvModule(
        (layer_norm): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
        (pw_conv1): Conv1d(512, 1024, kernel_size=(1,), stride=(1,))
        (glu): GLU(dim=1)
        (dw_conv): Conv1d(512, 512, kernel_size=(31,), stride=(1,), padding=(15,), groups=512)
        (bn): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (swish): Swish()
        (pw_conv2): Conv1d(512, 512, kernel_size=(1,), stride=(1,))
        (dropout): Dropout(p=0.1, inplace=False)
      )
      (ff2): FeedForwardModule(
        (layer_norm): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
        (linear1): Linear(in_features=512, out_features=2048, bias=True)
        (swish): Swish()
        (dropout): Dropout(p=0.1, inplace=False)
        (linear2): Linear(in_features=2048, out_features=512, bias=True)
        (dropout2): Dropout(p=0.1, inplace=False)
      )
      (final_ln): LayerNorm((512,), eps=1e-05, elementwise_affine=True)
    )
  )
  (ctc_head): Linear(in_features=512, out_features=4001, bias=True)
)
All parameters: 133758881
Trainable parameters: 133758881
2025-10-16 05:54:40,782 - torchaudio.utils.download - INFO - The local file (/home/jovyan/.cache/torch/hub/torchaudio/decoder-assets/librispeech-4-gram/lexicon.txt) exists. Skipping the download.
2025-10-16 05:54:40,784 - torchaudio.utils.download - INFO - The local file (/home/jovyan/.cache/torch/hub/torchaudio/decoder-assets/librispeech-4-gram/tokens.txt) exists. Skipping the download.
2025-10-16 05:54:40,785 - torchaudio.utils.download - INFO - The local file (/home/jovyan/.cache/torch/hub/torchaudio/decoder-assets/librispeech-4-gram/lm.bin) exists. Skipping the download.
2025-10-16 05:54:41,081 - pyctcdecode.decoder - WARNING - Unigrams not provided and cannot be automatically determined from LM file (only arpa format). Decoding accuracy might be reduced.
2025-10-16 05:54:41,082 - pyctcdecode.alphabet - INFO - Alphabet determined to be of BPE style.
2025-10-16 05:54:41,084 - pyctcdecode.alphabet - INFO - Found <unk> in vocabulary, substituting with ▁⁇▁.
2025-10-16 05:54:41,084 - pyctcdecode.language_model - WARNING - No known unigrams provided, decoding results might be a lot worse.
2025-10-16 05:54:41,088 - torchaudio.utils.download - INFO - The local file (/home/jovyan/.cache/torch/hub/torchaudio/decoder-assets/librispeech-4-gram/lexicon.txt) exists. Skipping the download.
2025-10-16 05:54:41,088 - torchaudio.utils.download - INFO - The local file (/home/jovyan/.cache/torch/hub/torchaudio/decoder-assets/librispeech-4-gram/tokens.txt) exists. Skipping the download.
2025-10-16 05:54:41,089 - torchaudio.utils.download - INFO - The local file (/home/jovyan/.cache/torch/hub/torchaudio/decoder-assets/librispeech-4-gram/lm.bin) exists. Skipping the download.
2025-10-16 05:54:41,336 - pyctcdecode.decoder - WARNING - Unigrams not provided and cannot be automatically determined from LM file (only arpa format). Decoding accuracy might be reduced.
2025-10-16 05:54:41,338 - pyctcdecode.alphabet - INFO - Alphabet determined to be of BPE style.
2025-10-16 05:54:41,340 - pyctcdecode.alphabet - INFO - Found <unk> in vocabulary, substituting with ▁⁇▁.
2025-10-16 05:54:41,340 - pyctcdecode.language_model - WARNING - No known unigrams provided, decoding results might be a lot worse.
2025-10-16 06:10:32,268 - train - INFO -     epoch          : 1
2025-10-16 06:10:32,270 - train - INFO -     loss           : 6.484967002868652
2025-10-16 06:10:32,271 - train - INFO -     grad_norm      : 1.9687985515594482
2025-10-16 06:10:32,272 - train - INFO -     val_loss       : 6.557441499811658
2025-10-16 06:10:32,273 - train - INFO -     val_CER_(Argmax): 1.0
2025-10-16 06:10:32,274 - train - INFO -     val_WER_(Argmax): 1.0
2025-10-16 06:10:32,275 - train - INFO -     test_loss      : 6.5487777634364805
2025-10-16 06:10:32,276 - train - INFO -     test_CER_(Argmax): 1.0
2025-10-16 06:10:32,277 - train - INFO -     test_WER_(Argmax): 1.0
2025-10-16 06:10:35,207 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 06:29:06,374 - train - INFO -     epoch          : 2
2025-10-16 06:29:06,376 - train - INFO -     loss           : 4.268152170181274
2025-10-16 06:29:06,377 - train - INFO -     grad_norm      : 2.9024139738082884
2025-10-16 06:29:06,378 - train - INFO -     val_loss       : 4.269274281326836
2025-10-16 06:29:06,379 - train - INFO -     val_CER_(Argmax): 0.6254875798647971
2025-10-16 06:29:06,380 - train - INFO -     val_WER_(Argmax): 0.7777343786937407
2025-10-16 06:29:06,381 - train - INFO -     test_loss      : 4.2218141235956335
2025-10-16 06:29:06,381 - train - INFO -     test_CER_(Argmax): 0.6175244368536745
2025-10-16 06:29:06,382 - train - INFO -     test_WER_(Argmax): 0.7649508240659034
2025-10-16 06:29:09,397 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 06:47:24,767 - train - INFO -     epoch          : 3
2025-10-16 06:47:24,769 - train - INFO -     loss           : 2.838803277015686
2025-10-16 06:47:24,770 - train - INFO -     grad_norm      : 3.8511923122406007
2025-10-16 06:47:24,771 - train - INFO -     val_loss       : 3.151403150614902
2025-10-16 06:47:24,772 - train - INFO -     val_CER_(Argmax): 0.5166045467920652
2025-10-16 06:47:24,773 - train - INFO -     val_WER_(Argmax): 0.6390588393546393
2025-10-16 06:47:24,774 - train - INFO -     test_loss      : 3.087251373907415
2025-10-16 06:47:24,775 - train - INFO -     test_CER_(Argmax): 0.5054364771394182
2025-10-16 06:47:24,776 - train - INFO -     test_WER_(Argmax): 0.6240195606539445
2025-10-16 06:47:27,893 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 07:05:26,375 - train - INFO -     epoch          : 4
2025-10-16 07:05:26,377 - train - INFO -     loss           : 2.192830967903137
2025-10-16 07:05:26,378 - train - INFO -     grad_norm      : 4.998111395835877
2025-10-16 07:05:26,379 - train - INFO -     val_loss       : 2.4318968276300374
2025-10-16 07:05:26,380 - train - INFO -     val_CER_(Argmax): 0.3702031224175337
2025-10-16 07:05:26,381 - train - INFO -     val_WER_(Argmax): 0.4860901782518249
2025-10-16 07:05:26,382 - train - INFO -     test_loss      : 2.3799445985293968
2025-10-16 07:05:26,383 - train - INFO -     test_CER_(Argmax): 0.36083352099915417
2025-10-16 07:05:26,383 - train - INFO -     test_WER_(Argmax): 0.4686573810151683
2025-10-16 07:05:29,594 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 07:23:08,170 - train - INFO -     epoch          : 5
2025-10-16 07:23:08,172 - train - INFO -     loss           : 1.7771303844451904
2025-10-16 07:23:08,173 - train - INFO -     grad_norm      : 3.3251154708862303
2025-10-16 07:23:08,174 - train - INFO -     val_loss       : 2.090497947303501
2025-10-16 07:23:08,175 - train - INFO -     val_CER_(Argmax): 0.3058225246754419
2025-10-16 07:23:08,176 - train - INFO -     val_WER_(Argmax): 0.41335918917225634
2025-10-16 07:23:08,177 - train - INFO -     test_loss      : 2.0416752507046954
2025-10-16 07:23:08,178 - train - INFO -     test_CER_(Argmax): 0.301632438606397
2025-10-16 07:23:08,179 - train - INFO -     test_WER_(Argmax): 0.40348729998697674
2025-10-16 07:23:11,300 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 07:40:48,673 - train - INFO -     epoch          : 6
2025-10-16 07:40:48,674 - train - INFO -     loss           : 1.5190469789505006
2025-10-16 07:40:48,675 - train - INFO -     grad_norm      : 4.879570946693421
2025-10-16 07:40:48,676 - train - INFO -     val_loss       : 1.8886281943180152
2025-10-16 07:40:48,677 - train - INFO -     val_CER_(Argmax): 0.2587877130236596
2025-10-16 07:40:48,678 - train - INFO -     val_WER_(Argmax): 0.3641306866341728
2025-10-16 07:40:48,679 - train - INFO -     test_loss      : 1.8447027896962516
2025-10-16 07:40:48,680 - train - INFO -     test_CER_(Argmax): 0.25522096907582587
2025-10-16 07:40:48,681 - train - INFO -     test_WER_(Argmax): 0.35652121745180754
2025-10-16 07:40:51,912 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 07:58:17,171 - train - INFO -     epoch          : 7
2025-10-16 07:58:17,173 - train - INFO -     loss           : 1.3213079261779785
2025-10-16 07:58:17,174 - train - INFO -     grad_norm      : 3.091304955482483
2025-10-16 07:58:17,175 - train - INFO -     val_loss       : 1.6745866332533796
2025-10-16 07:58:17,176 - train - INFO -     val_CER_(Argmax): 0.23459130517693363
2025-10-16 07:58:17,177 - train - INFO -     val_WER_(Argmax): 0.32632182944035326
2025-10-16 07:58:17,178 - train - INFO -     test_loss      : 1.6451512444310072
2025-10-16 07:58:17,179 - train - INFO -     test_CER_(Argmax): 0.2309938232961261
2025-10-16 07:58:17,179 - train - INFO -     test_WER_(Argmax): 0.31975252493656187
2025-10-16 07:58:20,480 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 08:15:34,076 - train - INFO -     epoch          : 8
2025-10-16 08:15:34,078 - train - INFO -     loss           : 1.183221834897995
2025-10-16 08:15:34,079 - train - INFO -     grad_norm      : 3.291339592933655
2025-10-16 08:15:34,080 - train - INFO -     val_loss       : 1.5990028621176997
2025-10-16 08:15:34,081 - train - INFO -     val_CER_(Argmax): 0.22179282085217464
2025-10-16 08:15:34,081 - train - INFO -     val_WER_(Argmax): 0.3144853777979456
2025-10-16 08:15:34,082 - train - INFO -     test_loss      : 1.5695159406196781
2025-10-16 08:15:34,083 - train - INFO -     test_CER_(Argmax): 0.21839530817241426
2025-10-16 08:15:34,084 - train - INFO -     test_WER_(Argmax): 0.30507432777976795
2025-10-16 08:15:37,270 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 08:32:47,776 - train - INFO -     epoch          : 9
2025-10-16 08:32:47,778 - train - INFO -     loss           : 1.0919397461414337
2025-10-16 08:32:47,779 - train - INFO -     grad_norm      : 3.284881491661072
2025-10-16 08:32:47,780 - train - INFO -     val_loss       : 1.572082411608047
2025-10-16 08:32:47,781 - train - INFO -     val_CER_(Argmax): 0.21293784639291272
2025-10-16 08:32:47,782 - train - INFO -     val_WER_(Argmax): 0.3055497990807289
2025-10-16 08:32:47,783 - train - INFO -     test_loss      : 1.5450149919928573
2025-10-16 08:32:47,784 - train - INFO -     test_CER_(Argmax): 0.21150730980404384
2025-10-16 08:32:47,784 - train - INFO -     test_WER_(Argmax): 0.3003562421297586
2025-10-16 08:32:50,854 - train - INFO - Saving current best: model_best.pth ...
2025-10-16 08:50:16,474 - train - INFO -     epoch          : 10
2025-10-16 08:50:16,476 - train - INFO -     loss           : 1.0653285849094392
2025-10-16 08:50:16,477 - train - INFO -     grad_norm      : 4.006628918647766
2025-10-16 08:50:16,478 - train - INFO -     val_loss       : 1.569695869846457
2025-10-16 08:50:16,479 - train - INFO -     val_CER_(Argmax): 0.21037508214070705
2025-10-16 08:50:16,480 - train - INFO -     val_WER_(Argmax): 0.3034500641559552
2025-10-16 08:50:16,481 - train - INFO -     test_loss      : 1.5422128947769724
2025-10-16 08:50:16,482 - train - INFO -     test_CER_(Argmax): 0.20823097080717934
2025-10-16 08:50:16,483 - train - INFO -     test_WER_(Argmax): 0.2974072927860874
2025-10-16 08:50:19,866 - train - INFO - Saving current best: model_best.pth ...