WhisperForSpokenNER-end2end
This model is a fine-tuned version of openai/whisper-small on the facebook/voxpopuli de+es+fr+nl dataset. It achieves the following results on the evaluation set:
- Loss: 0.3440
- Combined Wer: 0.2231
- F1 Score: 0.5368
- Label F1: 0.6908
- Wer: 0.1421
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: 64
- eval_batch_size: 64
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 500
- training_steps: 5000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Combined Wer | F1 Score | Label F1 | Wer |
|---|---|---|---|---|---|---|---|
| 1.1583 | 0.1 | 500 | 1.0361 | 0.3217 | 0.0746 | 0.1415 | 0.2067 |
| 0.4069 | 0.2 | 1000 | 0.4111 | 0.2203 | 0.4223 | 0.5940 | 0.1235 |
| 0.3708 | 0.3 | 1500 | 0.3768 | 0.2201 | 0.4609 | 0.6267 | 0.1295 |
| 0.3512 | 0.4 | 2000 | 0.3624 | 0.2223 | 0.5142 | 0.6835 | 0.1359 |
| 0.3411 | 0.5 | 2500 | 0.3543 | 0.2204 | 0.5225 | 0.6883 | 0.1374 |
| 0.3313 | 1.02 | 3000 | 0.3492 | 0.2235 | 0.5193 | 0.6808 | 0.1398 |
| 0.3252 | 1.12 | 3500 | 0.3459 | 0.2251 | 0.5333 | 0.6893 | 0.1436 |
| 0.3293 | 1.22 | 4000 | 0.3447 | 0.2237 | 0.5325 | 0.6860 | 0.1416 |
| 0.321 | 1.32 | 4500 | 0.3443 | 0.2238 | 0.5366 | 0.6905 | 0.1425 |
| 0.3223 | 1.42 | 5000 | 0.3440 | 0.2231 | 0.5368 | 0.6908 | 0.1421 |
Framework versions
- PEFT 0.7.1.dev0
- Transformers 4.37.0.dev0
- Pytorch 2.1.0
- Datasets 2.14.6
- Tokenizers 0.14.1