distilbert-base-uncased-IMDB_disbert1
This model is a fine-tuned version of distilbert-base-uncased on the None dataset. It achieves the following results on the evaluation set:
- Loss: 2.0461
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: 12
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.5838 | 1.0 | 10000 | 2.3985 |
| 2.4367 | 2.0 | 20000 | 2.3194 |
| 2.349 | 3.0 | 30000 | 2.2716 |
| 2.2764 | 4.0 | 40000 | 2.2392 |
| 2.2169 | 5.0 | 50000 | 2.1972 |
| 2.1623 | 6.0 | 60000 | 2.1655 |
| 2.122 | 7.0 | 70000 | 2.1277 |
| 2.0728 | 8.0 | 80000 | 2.1221 |
| 2.0348 | 9.0 | 90000 | 2.0928 |
| 1.9918 | 10.0 | 100000 | 2.0739 |
| 1.9621 | 11.0 | 110000 | 2.0536 |
| 1.9394 | 12.0 | 120000 | 2.0461 |
Framework versions
- Transformers 4.21.1
- Pytorch 1.12.1+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1