<!-- 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-vi-colab-all
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: inf
- Wer: 0.4537
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: 500
- num_epochs: 30
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
7.448 | 2.4 | 400 | inf | 1.0 |
2.8589 | 4.79 | 800 | inf | 0.7777 |
1.4919 | 7.19 | 1200 | inf | 0.5968 |
1.1255 | 9.58 | 1600 | inf | 0.5540 |
0.9354 | 11.98 | 2000 | inf | 0.4970 |
0.7816 | 14.37 | 2400 | inf | 0.4799 |
0.6822 | 16.77 | 2800 | inf | 0.4785 |
0.5768 | 19.16 | 3200 | inf | 0.4704 |
0.5031 | 21.56 | 3600 | inf | 0.4609 |
0.4589 | 23.95 | 4000 | inf | 0.4585 |
0.4136 | 26.35 | 4400 | inf | 0.4592 |
0.3829 | 28.74 | 4800 | inf | 0.4537 |
Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu113
- Datasets 1.18.3
- Tokenizers 0.10.3