mdeberta-targin-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: 0.5637
- Accuracy: 0.7091
- Precision: 0.6841
- Recall: 0.6557
- F1: 0.6617
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: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Precision | Recall | F1 |
|---|---|---|---|---|---|---|---|
| No log | 1.0 | 296 | 0.6001 | 0.6435 | 0.6344 | 0.5087 | 0.4156 |
| 0.6011 | 2.0 | 592 | 0.5633 | 0.7091 | 0.6879 | 0.6464 | 0.6521 |
| 0.6011 | 3.0 | 888 | 0.5501 | 0.7234 | 0.6991 | 0.6841 | 0.6892 |
| 0.5401 | 4.0 | 1184 | 0.5558 | 0.7082 | 0.6818 | 0.6595 | 0.6652 |
| 0.5401 | 5.0 | 1480 | 0.5637 | 0.7091 | 0.6841 | 0.6557 | 0.6617 |
Framework versions
- Transformers 4.24.0.dev0
- Pytorch 1.11.0+cu102
- Datasets 2.6.1
- Tokenizers 0.13.1