xlm-roberta-hate-final
This model is a fine-tuned version of xlm-roberta-base on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4961
- Accuracy: 0.8061
- Precision: 0.8145
- Recall: 0.8061
- F1: 0.7997
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 | 0.4747 | 0.7652 | 0.8023 | 0.7652 | 0.7456 |
| 0.5446 | 2.0 | 592 | 0.4620 | 0.7918 | 0.8183 | 0.7918 | 0.7789 |
| 0.5446 | 3.0 | 888 | 0.4580 | 0.8042 | 0.8144 | 0.8042 | 0.7971 |
| 0.453 | 4.0 | 1184 | 0.4534 | 0.7985 | 0.8071 | 0.7985 | 0.7915 |
| 0.453 | 5.0 | 1480 | 0.4749 | 0.8051 | 0.8147 | 0.8051 | 0.7983 |
| 0.406 | 6.0 | 1776 | 0.4961 | 0.8061 | 0.8145 | 0.8061 | 0.7997 |
Framework versions
- Transformers 4.24.0.dev0
- Pytorch 1.11.0+cu102
- Datasets 2.6.1
- Tokenizers 0.13.1