metadata
license: mit
tags:
- generated_from_trainer
datasets:
- squad
model-index:
- name: bert-tiny-finetuned-squad
results: []
bert-tiny-finetuned-squad
This model is a fine-tuned version of prajjwal1/bert-tiny on the squad dataset. It achieves the following results on the evaluation set:
- Loss: 2.3478
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: 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: 20
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.7198 | 1.0 | 2767 | 2.5916 |
| 2.6502 | 2.0 | 5534 | 2.5309 |
| 2.5589 | 3.0 | 8301 | 2.4849 |
| 2.5128 | 4.0 | 11068 | 2.4474 |
| 2.4631 | 5.0 | 13835 | 2.4226 |
| 2.4165 | 6.0 | 16602 | 2.3987 |
| 2.3634 | 7.0 | 19369 | 2.4002 |
| 2.3243 | 8.0 | 22136 | 2.3855 |
| 2.2799 | 9.0 | 24903 | 2.3827 |
| 2.2596 | 10.0 | 27670 | 2.3641 |
| 2.2133 | 11.0 | 30437 | 2.3464 |
| 2.1911 | 12.0 | 33204 | 2.3551 |
| 2.1621 | 13.0 | 35971 | 2.3553 |
| 2.1421 | 14.0 | 38738 | 2.3584 |
| 2.1202 | 15.0 | 41505 | 2.3501 |
| 2.1154 | 16.0 | 44272 | 2.3548 |
| 2.1237 | 17.0 | 47039 | 2.3456 |
| 2.0952 | 18.0 | 49806 | 2.3507 |
| 2.0886 | 19.0 | 52573 | 2.3500 |
| 2.1062 | 20.0 | 55340 | 2.3478 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.0
- Tokenizers 0.10.3