distilbert-hate-final
This model is a fine-tuned version of distilbert-base-multilingual-cased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.6212
- Accuracy: 0.7253
- Precision: 0.7207
- Recall: 0.7253
- F1: 0.7206
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: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 6
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall | F1 |
|---|---|---|---|---|---|---|---|
| No log | 1.0 | 296 | 0.5760 | 0.7025 | 0.7053 | 0.7025 | 0.6771 |
| 0.569 | 2.0 | 592 | 0.5629 | 0.7215 | 0.7168 | 0.7215 | 0.7122 |
| 0.569 | 3.0 | 888 | 0.5616 | 0.7310 | 0.7274 | 0.7310 | 0.7215 |
| 0.4683 | 4.0 | 1184 | 0.5651 | 0.7338 | 0.7295 | 0.7338 | 0.7274 |
| 0.4683 | 5.0 | 1480 | 0.5898 | 0.7338 | 0.7305 | 0.7338 | 0.7246 |
| 0.4086 | 6.0 | 1776 | 0.6212 | 0.7253 | 0.7207 | 0.7253 | 0.7206 |
Framework versions
- Transformers 4.24.0.dev0
- Pytorch 1.11.0+cu102
- Datasets 2.6.1
- Tokenizers 0.13.1