Edit model card

opus-mt-zh-de-tuned-Tatoeba-small

This model is a fine-tuned version of Helsinki-NLP/opus-mt-zh-de on a refined dataset of Tatoeba German - Chinese corpus https://github.com/Helsinki-NLP/Tatoeba-Challenge/blob/master/data/README.md. It achieves the following results on the evaluation set:

  • Loss: 2.2703
  • Bleu: 16.504
  • Gen Len: 16.6531

Model description

More information needed

Intended uses & limitations

Prefix used during fine-tuning: "将中文翻译成德语". This prefix is also recommended in prediction.

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: 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: 2
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Bleu Gen Len
2.7229 0.24 16000 2.5605 14.1956 16.2206
2.5988 0.49 32000 2.4447 14.8619 16.2726
2.515 0.73 48000 2.3817 15.3212 16.2823
2.4683 0.97 64000 2.3367 15.9043 16.7138
2.3873 1.22 80000 2.3115 16.1037 16.6369
2.3792 1.46 96000 2.2919 16.2957 16.6304
2.3626 1.7 112000 2.2790 16.2995 16.6235
2.3353 1.95 128000 2.2703 16.504 16.6531

Framework versions

  • Transformers 4.15.0
  • Pytorch 1.10.0+cu111
  • Datasets 1.17.0
  • Tokenizers 0.10.3
Downloads last month
2
Hosted inference API
This model can be loaded on the Inference API on-demand.

Evaluation results