Whisper Large-V3 Portuguese
This model is a fine-tuned version of openai/whisper-large-v3 on the mozilla-foundation/common_voice_13_0 pt dataset. It achieves the following results on the evaluation set:
- Loss: 0.4315
- Wer: 4.6003
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-06
- train_batch_size: 32
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 20000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0702 | 3.53 | 1000 | 0.1289 | 4.0367 |
| 0.0247 | 7.05 | 2000 | 0.1806 | 4.4294 |
| 0.0074 | 10.58 | 3000 | 0.2821 | 4.7481 |
| 0.0022 | 14.11 | 4000 | 0.3160 | 4.6249 |
| 0.0016 | 17.64 | 5000 | 0.3261 | 4.6479 |
| 0.0027 | 21.16 | 6000 | 0.3373 | 4.6479 |
| 0.0009 | 24.69 | 7000 | 0.3642 | 4.7087 |
| 0.0007 | 28.22 | 8000 | 0.3551 | 4.6611 |
| 0.0006 | 31.75 | 9000 | 0.3741 | 4.7481 |
| 0.0004 | 35.27 | 10000 | 0.3755 | 4.6791 |
| 0.0008 | 38.8 | 11000 | 0.3690 | 4.6381 |
| 0.0002 | 42.33 | 12000 | 0.3888 | 4.5115 |
| 0.0002 | 45.86 | 13000 | 0.3982 | 4.5855 |
| 0.0001 | 49.38 | 14000 | 0.4040 | 4.6085 |
| 0.0001 | 52.91 | 15000 | 0.4100 | 4.5888 |
| 0.0001 | 56.44 | 16000 | 0.4165 | 4.5871 |
| 0.0001 | 59.96 | 17000 | 0.4211 | 4.5855 |
| 0.0001 | 63.49 | 18000 | 0.4265 | 4.5838 |
| 0.0001 | 67.02 | 19000 | 0.4302 | 4.5921 |
| 0.0001 | 70.55 | 20000 | 0.4315 | 4.6003 |
Framework versions
- Transformers 4.37.2
- Pytorch 2.0.1+cu117
- Datasets 2.14.4
- Tokenizers 0.15.1