language:
- sw
- en
tags:
- translation
license: cc-by-4.0
HPLT MT release v1.0
This repository contains the translation model for sw-en trained with OPUS and HPLT data. For usage instructions, evaluation scripts, and inference scripts, please refer to the HPLT-MT-Models v1.0 GitHub repository.
Model Info
- Source language: Swahili
- Target language: English
- Dataset: OPUS and HPLT data
- Model architecture: Transformer-base
- Tokenizer: SentencePiece (Unigram)
- Cleaning: We used OpusCleaner with a set of basic rules. Details can be found in the filter files in Github
You can also read our deliverable report here for more details.
Usage
The model has been trained with Marian. To run inference, refer to the Inference/Decoding/Translation section of our GitHub repository.
The model can be used with the Hugging Face framework if the weights are converted to the Hugging Face format. We might provide this in the future; contributions are also welcome.
Benchmarks
| testset | BLEU | chrF++ | COMET22 |
|---|---|---|---|
| flores200 | 38.2 | 60.0 | 0.8249 |
| ntrex | 37.1 | 58.1 | 0.8267 |
Acknowledgements
This project has received funding from the European Union's Horizon Europe research and innovation programme under grant agreement No 101070350 and from UK Research and Innovation (UKRI) under the UK government's Horizon Europe funding guarantee [grant number 10052546]
Brought to you by researchers from the University of Edinburgh, Charles University in Prague, and the whole HPLT consortium.