<!-- 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. -->
wav2vec2-large-xls-r-300m-j-phoneme-colab-3
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_10_0 dataset. It achieves the following results on the evaluation set:
- Loss: 0.6478
- Wer: 0.3336
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: 2
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 40
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| No log | 1.0 | 397 | 1.0586 | 0.9425 |
| No log | 2.0 | 794 | 0.5773 | 0.5847 |
| 1.9827 | 3.0 | 1191 | 0.5243 | 0.4882 |
| 1.9827 | 4.0 | 1588 | 0.4735 | 0.4624 |
| 1.9827 | 5.0 | 1985 | 0.4967 | 0.4789 |
| 0.6004 | 6.0 | 2382 | 0.4703 | 0.4246 |
| 0.6004 | 7.0 | 2779 | 0.4555 | 0.4194 |
| 0.4911 | 8.0 | 3176 | 0.4692 | 0.4284 |
| 0.4911 | 9.0 | 3573 | 0.4589 | 0.3997 |
| 0.4911 | 10.0 | 3970 | 0.4988 | 0.4286 |
| 0.4275 | 11.0 | 4367 | 0.4851 | 0.4153 |
| 0.4275 | 12.0 | 4764 | 0.5020 | 0.4039 |
| 0.3784 | 13.0 | 5161 | 0.5491 | 0.4169 |
| 0.3784 | 14.0 | 5558 | 0.5211 | 0.4080 |
| 0.3784 | 15.0 | 5955 | 0.5124 | 0.3950 |
| 0.3362 | 16.0 | 6352 | 0.5121 | 0.3909 |
| 0.3362 | 17.0 | 6749 | 0.5503 | 0.3728 |
| 0.3046 | 18.0 | 7146 | 0.5363 | 0.3915 |
| 0.3046 | 19.0 | 7543 | 0.6112 | 0.4076 |
| 0.3046 | 20.0 | 7940 | 0.5884 | 0.3755 |
| 0.2785 | 21.0 | 8337 | 0.5639 | 0.3793 |
| 0.2785 | 22.0 | 8734 | 0.6246 | 0.3742 |
| 0.2513 | 23.0 | 9131 | 0.6014 | 0.3714 |
| 0.2513 | 24.0 | 9528 | 0.6195 | 0.3697 |
| 0.2513 | 25.0 | 9925 | 0.6004 | 0.3729 |
| 0.2296 | 26.0 | 10322 | 0.5793 | 0.3585 |
| 0.2296 | 27.0 | 10719 | 0.6178 | 0.3628 |
| 0.2114 | 28.0 | 11116 | 0.5974 | 0.3507 |
| 0.2114 | 29.0 | 11513 | 0.6056 | 0.3432 |
| 0.2114 | 30.0 | 11910 | 0.6190 | 0.3536 |
| 0.1944 | 31.0 | 12307 | 0.6293 | 0.3550 |
| 0.1944 | 32.0 | 12704 | 0.6236 | 0.3535 |
| 0.1777 | 33.0 | 13101 | 0.6456 | 0.3503 |
| 0.1777 | 34.0 | 13498 | 0.6629 | 0.3444 |
| 0.1777 | 35.0 | 13895 | 0.6585 | 0.3432 |
| 0.1644 | 36.0 | 14292 | 0.6528 | 0.3455 |
| 0.1644 | 37.0 | 14689 | 0.6460 | 0.3437 |
| 0.1521 | 38.0 | 15086 | 0.6441 | 0.3360 |
| 0.1521 | 39.0 | 15483 | 0.6531 | 0.3350 |
| 0.1521 | 40.0 | 15880 | 0.6478 | 0.3336 |
Framework versions
- Transformers 4.21.3
- Pytorch 1.10.0+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1
