metadata
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- emotion
model-index:
- name: nlp_bert_emo_classifier
results: []
nlp_bert_emo_classifier
This model is a fine-tuned version of bert-base-uncased on the emotion dataset. It achieves the following results on the evaluation set:
- Loss: 0.2791
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: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 4
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 0.8887 | 0.25 | 500 | 0.4212 |
| 0.3216 | 0.5 | 1000 | 0.3192 |
| 0.2649 | 0.75 | 1500 | 0.2746 |
| 0.2535 | 1.0 | 2000 | 0.2573 |
| 0.163 | 1.25 | 2500 | 0.2157 |
| 0.1868 | 1.5 | 3000 | 0.2118 |
| 0.1258 | 1.75 | 3500 | 0.2319 |
| 0.1726 | 2.0 | 4000 | 0.1853 |
| 0.1035 | 2.25 | 4500 | 0.2146 |
| 0.1135 | 2.5 | 5000 | 0.2207 |
| 0.1117 | 2.75 | 5500 | 0.2496 |
| 0.1145 | 3.0 | 6000 | 0.2482 |
| 0.0726 | 3.25 | 6500 | 0.2654 |
| 0.0828 | 3.5 | 7000 | 0.2622 |
| 0.0817 | 3.75 | 7500 | 0.2775 |
| 0.0689 | 4.0 | 8000 | 0.2791 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.12.0+cu113
- Datasets 2.4.0
- Tokenizers 0.10.3