barbaroo/nllb_200_1.3B_en_fo

Model Description

• Model Architecture: This model is based on the NLLB 1.3B architecture and weights.
• Languages: This checkpoint is fine-tuned to translate from English (en) to Faroese (fo).
• Size: ~1.3B parameters.
• Finetuning Datasets:
  • Sprotin_parallel
  • fo_en_synthetic
  • Training Regime: Trained until convergence (about 2 epochs).
• License: Inherits the original licenses of the NLLB 1.3B model.

Intended Use

• Primary Use Case: Translate text from English to Faroese.
• Audience: Researchers, developers, or anyone interested in Faroese language processing.
• Usage Scenarios:
  • Building Faroese-English translation tools
  • Language research and corpus analysis
  • Synthetic data creation

Important: While the model can produce fluent translations, it is not guaranteed to be perfectly accurate on all inputs. Users should verify critical or sensitive content through human experts.

Metrics

• Model performance measures:
  This model was evaluated using BLEU, chrF and BERT-score —metrics widely adopted by the machine translation community. Additionally, human evaluation was performed by two human experts using the ESA framework on a small dataset (about 200 sentences) of English sentences from news outlets (BBC, CNN, Al Jazeera).

Evaluation Data

• Datasets:
  Flores-200 dataset is described in Section 4 of the NLLB paper/documentation.
• Motivation:
  Flores-200 is currently the only machine translation benchmark available for Faroese.

  How to Use

Citation

Contact

• Primary Contact: < Barbara Scalvini/ barbaras@setur.fo / barbaralongview@gmail.com >