metadata
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: wav2vec2 russian
results: []
wav2vec2-russian
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: 1.2972
- Wer: 0.4920
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: 1e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 6
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.6885 | 0.25 | 500 | 1.2398 | 0.4926 |
| 0.6499 | 0.5 | 1000 | 1.2338 | 0.4957 |
| 0.6446 | 0.74 | 1500 | 1.2716 | 0.4975 |
| 0.6966 | 0.99 | 2000 | 1.5439 | 0.4942 |
| 0.7188 | 1.24 | 2500 | 1.4889 | 0.4960 |
| 0.7116 | 1.49 | 3000 | 1.0075 | 0.4921 |
| 0.7178 | 1.73 | 3500 | 1.3123 | 0.4984 |
| 0.7034 | 1.98 | 4000 | 1.1037 | 0.4919 |
| 0.7116 | 2.23 | 4500 | 0.9947 | 0.4942 |
| 0.7203 | 2.48 | 5000 | 1.1547 | 0.4940 |
| 0.72 | 2.73 | 5500 | 1.1245 | 0.4960 |
| 0.6921 | 2.97 | 6000 | 1.0844 | 0.4947 |
| 0.702 | 3.22 | 6500 | 1.1999 | 0.4931 |
| 0.6965 | 3.47 | 7000 | 1.2106 | 0.4906 |
| 0.6928 | 3.72 | 7500 | 1.1275 | 0.4939 |
| 0.6973 | 3.96 | 8000 | 1.1953 | 0.4922 |
| 0.7065 | 4.21 | 8500 | 1.2046 | 0.4931 |
| 0.6921 | 4.46 | 9000 | 1.2052 | 0.4926 |
| 0.6977 | 4.71 | 9500 | 1.2806 | 0.4922 |
| 0.688 | 4.96 | 10000 | 1.3327 | 0.4926 |
| 0.7003 | 5.2 | 10500 | 1.3518 | 0.4930 |
| 0.6981 | 5.45 | 11000 | 1.3295 | 0.4936 |
| 0.6913 | 5.7 | 11500 | 1.3065 | 0.4928 |
| 0.6822 | 5.95 | 12000 | 1.2972 | 0.4920 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.10.0+cu111
- Datasets 2.0.0
- Tokenizers 0.11.6