Edit model card

opus-mt-de-en-finetuned-de-to-en

This model is a fine-tuned version of Helsinki-NLP/opus-mt-de-en on the wmt14 dataset. It achieves the following results on the evaluation set:

  • eval_loss: 1.3411
  • eval_bleu: 32.4395
  • eval_gen_len: 29.6925
  • eval_runtime: 2250.0489
  • eval_samples_per_second: 19.998
  • eval_steps_per_second: 0.625
  • epoch: 3.0
  • step: 4221

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-06
  • train_batch_size: 32
  • eval_batch_size: 32
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 11
  • mixed_precision_training: Native AMP

Framework versions

  • Transformers 4.19.2
  • Pytorch 1.7.1+cu110
  • Datasets 2.2.2
  • Tokenizers 0.12.1
Downloads last month
0
Hosted inference API
This model can be loaded on the Inference API on-demand.

Dataset used to train PontifexMaximus/opus-mt-de-en-finetuned-de-to-en