File size: 2,148 Bytes
9553e70 61336ba 9553e70 61336ba 9553e70 61336ba 7f501e0 9553e70 92674bf 61336ba 7f501e0 9553e70 61336ba 7f501e0 61336ba 7f501e0 9553e70 7f501e0 9553e70 61336ba 9553e70 7f501e0 9553e70 61336ba 7f501e0 ca75df5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 |
---
language:
- en
- hr
tags:
- translation
license: cc-by-4.0
---
### HPLT MT release v1.0
This repository contains the translation model for en-hr trained with HPLT data only. For usage instructions, evaluation scripts, and inference scripts, please refer to the [HPLT-MT-Models v1.0](https://github.com/hplt-project/HPLT-MT-Models/tree/main/v1.0) GitHub repository.
### Model Info
* Source language: English
* Target language: Croatian
* Data: HPLT data only
* 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](https://github.com/hplt-project/HPLT-MT-Models/tree/main/v1.0/data/en-hr/raw/v0)
You can also read our deliverable report [here](https://hplt-project.org/HPLT_D5_1___Translation_models_for_select_language_pairs.pdf) for more details.
### Usage
**Note** that for quality considerations, we recommend using [HPLT/translate-en-hr-v1.0-hplt_opus](https://huggingface.co/HPLT/translate-en-hr-v1.0-hplt_opus) instead of this model.
The model has been trained with Marian. To run inference, refer to the [Inference/Decoding/Translation](https://github.com/hplt-project/HPLT-MT-Models/tree/main/v1.0#inferencedecodingtranslation) 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 | 28.4 | 54.9 | 0.8664 |
| ntrex | 28.3 | 53.9 | 0.8092 |
### 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.
|