albert-large-v2-spoken-squad
This model is a fine-tuned version of albert-large-v2 on the Spoken Squad dataset. It achieves the following results on the evaluation set:
- Exact Match: 66.7026
- F1: 79.3491
- Loss: 1.0481
Model description
Results on Spoken Squad Test Sets
| Test Set | Test Loss | Samples | Exact Match | F1 |
|---|---|---|---|---|
| Test | 1.183 | 5351 | 71.2951 | 80.4348 |
| Test WER44 | 6.2158 | 5351 | 45.9727 | 60.8491 |
| Test WER54 | 6.2158 | 5351 | 45.9727 | 60.8491 |
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: 2e-05
- train_batch_size: 16
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Exact Match | F1 | Validation Loss |
|---|---|---|---|---|---|
| 1.0444 | 1.0 | 2088 | 63.6584 | 77.0975 | 1.0645 |
| 0.8017 | 2.0 | 4176 | 66.3524 | 79.3253 | 0.9756 |
| 0.5426 | 3.0 | 6264 | 66.7026 | 79.3491 | 1.0481 |
Framework versions
- Transformers 4.24.0
- Pytorch 1.13.1
- Datasets 2.8.0
- Tokenizers 0.11.0