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