distilbert-sentiment-new
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: 0.5872
- Accuracy: 0.7243
- Precision: 0.7192
- Recall: 0.7243
- F1: 0.7175
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 | 0.6038 | 0.6787 | 0.7049 | 0.6787 | 0.6235 |
| 0.5926 | 2.0 | 592 | 0.5532 | 0.7148 | 0.7118 | 0.7148 | 0.6994 |
| 0.5926 | 3.0 | 888 | 0.5480 | 0.7243 | 0.7199 | 0.7243 | 0.7144 |
| 0.4946 | 4.0 | 1184 | 0.5535 | 0.7300 | 0.7255 | 0.7300 | 0.7220 |
| 0.4946 | 5.0 | 1480 | 0.5858 | 0.7186 | 0.7140 | 0.7186 | 0.7146 |
| 0.4267 | 6.0 | 1776 | 0.5872 | 0.7243 | 0.7192 | 0.7243 | 0.7175 |
Framework versions
- Transformers 4.24.0.dev0
- Pytorch 1.11.0+cu102
- Datasets 2.6.1
- Tokenizers 0.13.1