twitter-emotion-deberta-v3-base

This model is a fine-tuned version of DeBERTa-v3. It achieves the following results on the evaluation set:

• Loss: 0.1474
• Accuracy: 0.937

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 80
• eval_batch_size: 80
• lr_scheduler_type: linear
• num_epochs: 6.0

Framework versions

• Transformers 4.12.5
• Pytorch 1.10.0+cu113
• Datasets 1.15.1
• Tokenizers 0.10.3