twitter_RoBERTa_token_itr0_1e-05_webDiscourse_01_03_2022-14_57_21
This model is a fine-tuned version of cardiffnlp/twitter-roberta-base on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.5905
- Precision: 0.0024
- Recall: 0.0143
- F1: 0.0041
- Accuracy: 0.6867
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: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
|---|---|---|---|---|---|---|---|
| No log | 1.0 | 10 | 0.6081 | 0.0 | 0.0 | 0.0 | 0.6904 |
| No log | 2.0 | 20 | 0.6014 | 0.0025 | 0.0130 | 0.0042 | 0.6934 |
| No log | 3.0 | 30 | 0.5953 | 0.0 | 0.0 | 0.0 | 0.6930 |
| No log | 4.0 | 40 | 0.5858 | 0.0 | 0.0 | 0.0 | 0.6941 |
| No log | 5.0 | 50 | 0.5815 | 0.0 | 0.0 | 0.0 | 0.6947 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.10.1+cu113
- Datasets 1.18.0
- Tokenizers 0.10.3