bert-finetuned-protagonist
This model is a fine-tuned version of Davlan/bert-base-multilingual-cased-ner-hrl on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.0745
- Precision: 0.8392
- Recall: 0.7767
- F1: 0.8068
- Accuracy: 0.9863
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: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
|---|---|---|---|---|---|---|---|
| No log | 1.0 | 106 | 0.0695 | 0.8251 | 0.8558 | 0.8402 | 0.9870 |
| No log | 2.0 | 212 | 0.0667 | 0.8244 | 0.7860 | 0.8048 | 0.9857 |
| No log | 3.0 | 318 | 0.0624 | 0.86 | 0.8 | 0.8289 | 0.9870 |
| No log | 4.0 | 424 | 0.0698 | 0.8357 | 0.8047 | 0.8199 | 0.9867 |
| 0.0074 | 5.0 | 530 | 0.0745 | 0.8392 | 0.7767 | 0.8068 | 0.9863 |
Framework versions
- Transformers 4.18.0
- Pytorch 1.11.0+cu113
- Datasets 2.1.0
- Tokenizers 0.12.1