distilbert-finetuning-fakenews

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

• Loss: 0.2804
• Accuracy: 0.8833
• F1: 0.9014

Model description

More information needed

Intended uses & limitations

A DistilBERT model is trained on an external dataset (Spanish Fake and Real News) to detect fake news in spanish.

Training and evaluation data

Dataset obtained from: https://www.kaggle.com/datasets/zulanac/fake-and-real-news, under a CC BY-SA 4.0 license. Author: Fabricio A. Zules.

For compatibility reasons with the model, it was necessary to change 'texto' and 'clase' headers to 'text' and 'label'; and 'fake' and 'true' values (from class/label), were replaced by '0' and '1' values.

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: 3

Training results

Framework versions

• Transformers 4.24.0
• Pytorch 1.12.1+cu113
• Datasets 2.7.1
• Tokenizers 0.13.2