| 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 ... |
|
|
