mbart50-finetuned-multi30-en-to-de

This model is a fine-tuned version of facebook/mbart-large-50-one-to-many-mmt on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5946
• Bleu: 48.2650

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: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.13.0
• Pytorch 1.10.0+cu113
• Datasets 2.2.2
• Tokenizers 0.10.3