<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
bert-emotion
This model is a fine-tuned version of distilbert-base-cased on the tweet_eval dataset. It achieves the following results on the evaluation set:
- Loss: 1.2053
- Precision: 0.7072
- Recall: 0.7233
- Fscore: 0.7124
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: 4
- eval_batch_size: 4
- 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 | Precision | Recall | Fscore |
---|---|---|---|---|---|---|
0.8623 | 1.0 | 815 | 0.7198 | 0.7536 | 0.6312 | 0.6559 |
0.5637 | 2.0 | 1630 | 0.8756 | 0.7213 | 0.7166 | 0.7160 |
0.2845 | 3.0 | 2445 | 1.2053 | 0.7072 | 0.7233 | 0.7124 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2