distilbert-sst2

This model is a fine-tuned version of distilbert-base-uncased on the glue dataset. It achieves the following results on the evaluation set:

• eval_loss: 0.4182
• eval_accuracy: 0.8911
• eval_runtime: 1.8021
• eval_samples_per_second: 483.882
• eval_steps_per_second: 60.485
• epoch: 0.8
• step: 6700

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: 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: 3.0

Framework versions

• Transformers 4.15.0
• Pytorch 1.10.1
• Datasets 1.17.0
• Tokenizers 0.10.3