mbart-neutralization

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

• Loss: 0.0323
• Bleu: 98.8828
• Gen Len: 18.6771

Model description

This model translates from “standard” to neutral Spanish. For example, it replaces gender markers in plurals with neuter constructions.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5.6e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
No log	1.0	440	0.0149	98.9442	18.5938
0.2266	2.0	880	0.0323	98.8828	18.6771

Framework versions

• Transformers 4.50.1
• Pytorch 2.6.0+cu124
• Datasets 3.4.1
• Tokenizers 0.21.1