distilbert-emotion
This model is a fine-tuned version of distilbert-base-uncased on the emotion dataset. It achieves the following results on the evaluation set:
- Loss: 0.1333
- Accuracy: 0.9385
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: 5e-05
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| No log | 1.0 | 250 | 0.1954 | 0.926 |
| 0.3494 | 2.0 | 500 | 0.1472 | 0.937 |
| 0.3494 | 3.0 | 750 | 0.1333 | 0.9385 |
Framework versions
- Transformers 4.34.1
- Pytorch 2.1.0+cpu
- Datasets 2.14.5
- Tokenizers 0.14.1