File size: 2,147 Bytes
c1e6e20
 
 
 
 
 
 
 
 
132c3ea
c1e6e20
132c3ea
 
 
c1e6e20
c8ffbcd
 
132c3ea
 
c1e6e20
132c3ea
 
 
 
 
c8ffbcd
132c3ea
 
c1e6e20
132c3ea
c1e6e20
 
 
132c3ea
c1e6e20
f447f80
 
132c3ea
 
 
 
 
3e710d0
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
  - mt
tags:
- translation
license: cc-by-4.0
---

### HPLT MT release v1.0

This repository contains the translation model for en-mt 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: Maltese
* 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-mt/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-mt-v1.0-hplt_opus](https://huggingface.co/HPLT/translate-en-mt-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     | 30.6 | 60.7  | 0.6995  |
| ntrex | 24.1   | 54.3  | 0.6773  |

### 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.