bert-small-finetuned-squad
This model is a fine-tuned version of prajjwal1/bert-small on the squad dataset. It achieves the following results on the evaluation set:
- eval_loss: 1.3138
- eval_runtime: 46.6577
- eval_samples_per_second: 231.13
- eval_steps_per_second: 14.446
- epoch: 4.0
- step: 22132
{'exact_match': 71.05960264900662, 'f1': 80.8260245470904}
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: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 20
Framework versions
- Transformers 4.15.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.0
- Tokenizers 0.10.3