deberta-v3-base-finetuned-rte

This model is a fine-tuned version of microsoft/deberta-v3-base on the glue dataset. It achieves the following results on the evaluation set:

• Loss: 0.8234
• Accuracy: 0.8195

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	156	0.5610	0.7545
No log	2.0	312	0.6270	0.7617
No log	3.0	468	0.6565	0.7906
0.3919	4.0	624	0.8234	0.8195
0.3919	5.0	780	0.9628	0.7978

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu116
• Datasets 2.7.1
• Tokenizers 0.13.2