distilbert-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: 0.6307
- Accuracy: 0.6882
- Precision: 0.6443
- Recall: 0.6384
- F1: 0.6409
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 | 0.5882 | 0.6854 | 0.6355 | 0.6182 | 0.6226 |
| 0.5995 | 2.0 | 592 | 0.5693 | 0.7015 | 0.6590 | 0.6019 | 0.6030 |
| 0.5995 | 3.0 | 888 | 0.5823 | 0.6882 | 0.6440 | 0.6377 | 0.6403 |
| 0.5299 | 4.0 | 1184 | 0.5968 | 0.6949 | 0.6488 | 0.6340 | 0.6386 |
| 0.5299 | 5.0 | 1480 | 0.6236 | 0.6835 | 0.6430 | 0.6436 | 0.6433 |
| 0.4698 | 6.0 | 1776 | 0.6307 | 0.6882 | 0.6443 | 0.6384 | 0.6409 |
Framework versions
- Transformers 4.24.0.dev0
- Pytorch 1.11.0+cu102
- Datasets 2.6.1
- Tokenizers 0.13.1