scibert-cased-AstrID-tok
This model is a fine-tuned version of allenai/scibert_scivocab_cased on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.1843
- Precision: 0.7307
- Recall: 0.7575
- F1: 0.7439
- Accuracy: 0.9517
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: 3
Training results
| Training Loss | Epoch | Step | Validation Loss | Precision | Recall | F1 | Accuracy |
|---|---|---|---|---|---|---|---|
| No log | 1.0 | 220 | 0.3141 | 0.6254 | 0.6518 | 0.6383 | 0.9182 |
| No log | 2.0 | 440 | 0.1988 | 0.7002 | 0.7275 | 0.7136 | 0.9480 |
| 0.3514 | 3.0 | 660 | 0.1843 | 0.7307 | 0.7575 | 0.7439 | 0.9517 |
Framework versions
- Transformers 4.21.0
- Pytorch 1.12.0+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1