metadata
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- common_voice
model-index:
- name: xls-r-kyrgiz-cv8
results: []
xls-r-kyrgiz-cv8
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.5495
- Wer: 0.2951
- Cer: 0.0789
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: 32
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 300.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
|---|---|---|---|---|---|
| 3.1079 | 18.51 | 500 | 2.6795 | 0.9996 | 0.9825 |
| 0.8506 | 37.04 | 1000 | 0.4323 | 0.3718 | 0.0961 |
| 0.6821 | 55.55 | 1500 | 0.4105 | 0.3311 | 0.0878 |
| 0.6091 | 74.07 | 2000 | 0.4281 | 0.3168 | 0.0851 |
| 0.5429 | 92.58 | 2500 | 0.4525 | 0.3147 | 0.0842 |
| 0.5063 | 111.11 | 3000 | 0.4619 | 0.3144 | 0.0839 |
| 0.4661 | 129.62 | 3500 | 0.4660 | 0.3039 | 0.0818 |
| 0.4353 | 148.15 | 4000 | 0.4695 | 0.3083 | 0.0820 |
| 0.4048 | 166.65 | 4500 | 0.4909 | 0.3085 | 0.0824 |
| 0.3852 | 185.18 | 5000 | 0.5074 | 0.3048 | 0.0812 |
| 0.3567 | 203.69 | 5500 | 0.5111 | 0.3012 | 0.0810 |
| 0.3451 | 222.22 | 6000 | 0.5225 | 0.2982 | 0.0804 |
| 0.325 | 240.73 | 6500 | 0.5270 | 0.2955 | 0.0796 |
| 0.3089 | 259.25 | 7000 | 0.5381 | 0.2929 | 0.0793 |
| 0.2941 | 277.76 | 7500 | 0.5565 | 0.2923 | 0.0794 |
| 0.2945 | 296.29 | 8000 | 0.5495 | 0.2951 | 0.0789 |
Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.2+cu102
- Datasets 1.18.3
- Tokenizers 0.11.0