mbart-neutralization

This model is a fine-tuned version of facebook/mbart-large-50 on the dataset "somosnlp-hackathon-2022/neutral-es". It achieves the following results on the evaluation set:

• Loss: 0.0416
• Bleu: 96.855
• Gen Len: 18.5417

Model description

This model is designed to convert Spanish gendered text into inclusive language. It was developed as part of a Master’s degree project in Natural Language Processing (NLP).

Intended uses & limitations

The model was trained on a relatively small dataset, so its performance is limited and the results may not always be reliable. It is intended primarily for educational and experimental purposes.

Training and evaluation data

The model was trained with this dataset of Spanish Gender Neutralization: https://huggingface.co/datasets/somosnlp-hackathon-2022/neutral-es

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5.6e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Use OptimizerNames.ADAFACTOR and the args are: No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
0.0729	1.0	3513	0.0961	94.3227	18.3021
0.0357	2.0	7026	0.0416	96.855	18.5417

Framework versions

• Transformers 5.2.0
• Pytorch 2.10.0+cu128
• Datasets 4.6.0
• Tokenizers 0.22.2