dknews-NB-BERT-AI-classifier/

This model is a fine-tuned version of NbAiLab/nb-bert-large on a custom dataset with Danish News articles either generated by GPT-3 or a Danish journalist from a large Danish news media. The task is then to classify whether the article is written by GPT-3 (label = 0) or human (label = 1)

It achieves the following results on the evaluation set (the best model loaded i.e., after 2 epochs)

• Loss: 0.1804
• Accuracy: 0.9574
• F1: 0.9574
• Precision: 0.9576
• Recall: 0.9574

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

The model is trained on Danish news articles either generated by a fine-tuned GPT-3 or a Danish Journalist from a large Danish News Media TV2.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 24
• eval_batch_size: 24
• seed: 2502
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
0.696	1.0	39	0.4926	0.8262	0.8211	0.8672	0.8262
0.4195	2.0	78	0.1804	0.9574	0.9574	0.9576	0.9574
0.1458	3.0	117	0.2810	0.9246	0.9241	0.9344	0.9246
0.0424	4.0	156	0.5893	0.8852	0.8838	0.9041	0.8852
0.0246	5.0	195	1.4776	0.7475	0.7301	0.8321	0.7475

Framework versions

• Transformers 4.24.0
• Pytorch 1.12.1+cu113
• Datasets 2.7.0
• Tokenizers 0.13.2