toki-en-mt / README.md
lbourdois's picture
Add multilingual to the language tag
5ff218e
|
raw
history blame
2.21 kB
metadata
language:
  - tok
  - en
  - multilingual
license: apache-2.0
tags:
  - generated_from_trainer
  - translation
metrics:
  - bleu
widget:
  - text: toki! mi jan Ton. mi lon ma Tawan.
  - text: soweli li toki ala toki e toki Inli?
model-index:
  - name: toki-en-mt
    results: []

toki-en-mt

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

  • Loss: 1.2840
  • Bleu: 26.7612
  • Gen Len: 9.0631

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: 16
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 10
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Bleu Gen Len
1.7228 1.0 1260 1.4572 19.9464 9.2177
1.3182 2.0 2520 1.3356 22.4628 9.1263
1.1241 3.0 3780 1.3028 23.5152 9.0462
0.9995 4.0 5040 1.2784 23.9526 9.1697
0.8945 5.0 6300 1.2739 24.7707 9.0914
0.8331 6.0 7560 1.2725 25.3477 9.0518
0.7641 7.0 8820 1.2770 26.165 9.0245
0.7163 8.0 10080 1.2809 25.8053 9.0933
0.6886 9.0 11340 1.2799 26.5752 9.0669
0.6627 10.0 12600 1.2840 26.7612 9.0631

Framework versions

  • Transformers 4.20.1
  • Pytorch 1.11.0
  • Datasets 2.3.2
  • Tokenizers 0.12.1