<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
nick_asr_v2
This model is a fine-tuned version of ntoldalagi/nick_asr_v2 on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.4562
- Wer: 0.6422
- Cer: 0.2409
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 20
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Cer | Validation Loss | Wer |
|---|---|---|---|---|---|
| 0.2616 | 0.44 | 300 | 0.2905 | 1.2200 | 0.7496 |
| 0.441 | 0.87 | 600 | 0.2866 | 1.1936 | 0.7385 |
| 0.4366 | 1.31 | 900 | 0.2795 | 1.1584 | 0.7274 |
| 0.3982 | 1.75 | 1200 | 0.2808 | 1.2033 | 0.7274 |
| 0.3891 | 2.18 | 1500 | 0.2753 | 1.2044 | 0.7166 |
| 0.3508 | 2.91 | 2000 | 1.2382 | 0.7220 | 0.2743 |
| 0.2783 | 4.37 | 3000 | 1.3327 | 0.7177 | 0.2705 |
| 0.2495 | 5.82 | 4000 | 1.2286 | 0.6749 | 0.2638 |
| 0.1982 | 7.28 | 5000 | 1.3073 | 0.6721 | 0.2585 |
| 0.1717 | 8.73 | 6000 | 1.2941 | 0.6627 | 0.2500 |
| 0.1508 | 10.19 | 7000 | 1.3625 | 0.6584 | 0.2490 |
| 0.1329 | 11.64 | 8000 | 1.3863 | 0.6584 | 0.2474 |
| 0.1303 | 13.1 | 9000 | 1.3714 | 0.6534 | 0.2449 |
| 0.1159 | 14.56 | 10000 | 1.4043 | 0.6473 | 0.2442 |
| 0.1015 | 16.01 | 11000 | 1.4245 | 0.6498 | 0.2419 |
| 0.098 | 17.47 | 12000 | 1.4410 | 0.6440 | 0.2425 |
| 0.0869 | 18.92 | 13000 | 1.4562 | 0.6422 | 0.2409 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.0+cu111
- Datasets 2.0.0
- Tokenizers 0.12.1
