metadata
license: apache-2.0
tags:
- generated_from_trainer
metrics:
- wer
model-index:
- name: wav2vec2-large-asr-th
results: []
wav2vec2-large-asr-th
This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5920
- Wer: 0.5256
- Cer: 0.1778
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: 0.0001
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- training_steps: 6000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 3.6977 | 0.57 | 500 | 3.6087 | 1.0 | 0.9999 |
| 3.4771 | 1.14 | 1000 | 3.4975 | 1.0 | 0.9999 |
| 2.6942 | 1.71 | 1500 | 2.2851 | 1.0067 | 0.6597 |
| 1.698 | 2.28 | 2000 | 1.0650 | 0.7875 | 0.3045 |
| 1.5008 | 2.85 | 2500 | 0.8698 | 0.6838 | 0.2508 |
| 1.1706 | 3.42 | 3000 | 0.7382 | 0.6132 | 0.2140 |
| 1.1872 | 4.0 | 3500 | 0.6924 | 0.5840 | 0.2029 |
| 1.1422 | 4.57 | 4000 | 0.6531 | 0.5690 | 0.1959 |
| 0.9556 | 5.14 | 4500 | 0.6246 | 0.5432 | 0.1850 |
| 1.0091 | 5.71 | 5000 | 0.6052 | 0.5360 | 0.1822 |
| 1.0523 | 6.28 | 5500 | 0.5995 | 0.5293 | 0.1802 |
| 1.0205 | 6.85 | 6000 | 0.5920 | 0.5256 | 0.1778 |
Framework versions
- Transformers 4.26.1
- Pytorch 1.13.1+cu116
- Datasets 2.9.0
- Tokenizers 0.13.2