hasoc19-bert-base-multilingual-cased-sentiment-new
This model is a fine-tuned version of bert-base-multilingual-cased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4758
- Accuracy: 0.8501
- Precision: 0.8524
- Recall: 0.8501
- F1: 0.8507
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 |
|---|---|---|---|---|---|---|---|
| 0.4987 | 1.0 | 537 | 0.3931 | 0.8249 | 0.8304 | 0.8249 | 0.8208 |
| 0.3676 | 2.0 | 1074 | 0.3513 | 0.8417 | 0.8487 | 0.8417 | 0.8427 |
| 0.2883 | 3.0 | 1611 | 0.3604 | 0.8475 | 0.8484 | 0.8475 | 0.8478 |
| 0.235 | 4.0 | 2148 | 0.3929 | 0.8501 | 0.8509 | 0.8501 | 0.8504 |
| 0.1859 | 5.0 | 2685 | 0.4413 | 0.8449 | 0.8451 | 0.8449 | 0.8450 |
| 0.1563 | 6.0 | 3222 | 0.4758 | 0.8501 | 0.8524 | 0.8501 | 0.8507 |
Framework versions
- Transformers 4.24.0.dev0
- Pytorch 1.11.0+cu102
- Datasets 2.6.1
- Tokenizers 0.13.1