wav2vec-turkish-1b
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_16_1 dataset. It achieves the following results on the evaluation set:
- Loss: 0.2368
- Wer: 0.2462
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: 0.0003
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 0.1
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 1.9325 | 0.29 | 400 | 0.6568 | 0.7114 |
| 0.5559 | 0.58 | 800 | 0.4940 | 0.5587 |
| 0.4528 | 0.88 | 1200 | 0.4380 | 0.5008 |
| 0.3921 | 1.17 | 1600 | 0.3801 | 0.4465 |
| 0.3669 | 1.46 | 2000 | 0.3550 | 0.4367 |
| 0.3463 | 1.75 | 2400 | 0.3619 | 0.4440 |
| 0.323 | 2.05 | 2800 | 0.3476 | 0.4118 |
| 0.2872 | 2.34 | 3200 | 0.3208 | 0.3960 |
| 0.2836 | 2.63 | 3600 | 0.3069 | 0.3842 |
| 0.2728 | 2.92 | 4000 | 0.3072 | 0.3679 |
| 0.2486 | 3.21 | 4400 | 0.3287 | 0.3833 |
| 0.2347 | 3.51 | 4800 | 0.3266 | 0.3781 |
| 0.2309 | 3.8 | 5200 | 0.3005 | 0.3650 |
| 0.2202 | 4.09 | 5600 | 0.2916 | 0.3489 |
| 0.206 | 4.38 | 6000 | 0.2975 | 0.3390 |
| 0.1971 | 4.67 | 6400 | 0.2901 | 0.3377 |
| 0.1938 | 4.97 | 6800 | 0.2612 | 0.3188 |
| 0.1788 | 5.26 | 7200 | 0.2676 | 0.3183 |
| 0.169 | 5.55 | 7600 | 0.2691 | 0.3140 |
| 0.167 | 5.84 | 8000 | 0.2658 | 0.3099 |
| 0.1551 | 6.14 | 8400 | 0.2643 | 0.2986 |
| 0.1437 | 6.43 | 8800 | 0.2667 | 0.2998 |
| 0.1407 | 6.72 | 9200 | 0.2579 | 0.2955 |
| 0.1344 | 7.01 | 9600 | 0.2415 | 0.2820 |
| 0.122 | 7.3 | 10000 | 0.2476 | 0.2776 |
| 0.1205 | 7.6 | 10400 | 0.2555 | 0.2745 |
| 0.1169 | 7.89 | 10800 | 0.2400 | 0.2660 |
| 0.1071 | 8.18 | 11200 | 0.2537 | 0.2630 |
| 0.1041 | 8.47 | 11600 | 0.2466 | 0.2638 |
| 0.0994 | 8.77 | 12000 | 0.2413 | 0.2582 |
| 0.0925 | 9.06 | 12400 | 0.2340 | 0.2527 |
| 0.0877 | 9.35 | 12800 | 0.2393 | 0.2515 |
| 0.0859 | 9.64 | 13200 | 0.2418 | 0.2489 |
| 0.0847 | 9.93 | 13600 | 0.2368 | 0.2462 |
Framework versions
- Transformers 4.38.2
- Pytorch 2.2.2+cu121
- Datasets 2.17.1
- Tokenizers 0.15.2