correct_twitter_RoBERTa_token_itr0_1e-05_webDiscourse_01_03_2022-15_30_39
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: 0.6169
- Precision: 0.0031
- Recall: 0.0357
- F1: 0.0057
- Accuracy: 0.6464
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 | 0.6339 | 0.0116 | 0.0120 | 0.0118 | 0.6662 |
| No log | 2.0 | 20 | 0.6182 | 0.0064 | 0.0120 | 0.0084 | 0.6688 |
| No log | 3.0 | 30 | 0.6139 | 0.0029 | 0.0241 | 0.0052 | 0.6659 |
| No log | 4.0 | 40 | 0.6172 | 0.0020 | 0.0241 | 0.0037 | 0.6622 |
| No log | 5.0 | 50 | 0.6165 | 0.0019 | 0.0241 | 0.0036 | 0.6599 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.10.1+cu113
- Datasets 1.18.0
- Tokenizers 0.10.3