<!-- 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. -->
firstcolab
This model is a fine-tuned version of Thanapon1998/firstcolab on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.9090
- Wer: 0.9642
- Cer: 0.2127
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: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 2
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer | Cer |
---|---|---|---|---|---|
1.5413 | 0.66 | 100 | 1.1761 | 0.9887 | 0.2718 |
1.4052 | 1.32 | 200 | 1.0352 | 0.9809 | 0.2396 |
1.2922 | 1.97 | 300 | 0.9090 | 0.9642 | 0.2127 |
Framework versions
- Transformers 4.24.0
- Pytorch 1.12.1+cu113
- Datasets 2.7.1
- Tokenizers 0.13.2