File size: 8,907 Bytes
47304ea
 
 
 
cec8975
47304ea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cec8975
47304ea
cec8975
 
47304ea
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cec8975
47304ea
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
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
---
language:
- da
- he
- sv

tags:
- translation
- opus-mt-tc

license: cc-by-4.0
model-index:
- name: opus-mt-tc-big-gmq-he
  results:
  - task:
      name: Translation dan-heb
      type: translation
      args: dan-heb
    dataset:
      name: flores101-devtest
      type: flores_101
      args: dan heb devtest
    metrics:
       - name: BLEU
         type: bleu
         value: 22.9
       - name: chr-F
         type: chrf
         value: 0.52815
  - task:
      name: Translation isl-heb
      type: translation
      args: isl-heb
    dataset:
      name: flores101-devtest
      type: flores_101
      args: isl heb devtest
    metrics:
       - name: BLEU
         type: bleu
         value: 14.2
       - name: chr-F
         type: chrf
         value: 0.42284
  - task:
      name: Translation nob-heb
      type: translation
      args: nob-heb
    dataset:
      name: flores101-devtest
      type: flores_101
      args: nob heb devtest
    metrics:
       - name: BLEU
         type: bleu
         value: 19.2
       - name: chr-F
         type: chrf
         value: 0.49492
  - task:
      name: Translation swe-heb
      type: translation
      args: swe-heb
    dataset:
      name: flores101-devtest
      type: flores_101
      args: swe heb devtest
    metrics:
       - name: BLEU
         type: bleu
         value: 23.0
       - name: chr-F
         type: chrf
         value: 0.52408
---
# opus-mt-tc-big-gmq-he

## Table of Contents
- [Model Details](#model-details)
- [Uses](#uses)
- [Risks, Limitations and Biases](#risks-limitations-and-biases)
- [How to Get Started With the Model](#how-to-get-started-with-the-model)
- [Training](#training)
- [Evaluation](#evaluation)
- [Citation Information](#citation-information)
- [Acknowledgements](#acknowledgements)

## Model Details

Neural machine translation model for translating from North Germanic languages (gmq) to Hebrew (he).

This model is part of the [OPUS-MT project](https://github.com/Helsinki-NLP/Opus-MT), an effort to make neural machine translation models widely available and accessible for many languages in the world. All models are originally trained using the amazing framework of [Marian NMT](https://marian-nmt.github.io/), an efficient NMT implementation written in pure C++. The models have been converted to pyTorch using the transformers library by huggingface. Training data is taken from [OPUS](https://opus.nlpl.eu/) and training pipelines use the procedures of [OPUS-MT-train](https://github.com/Helsinki-NLP/Opus-MT-train).
**Model Description:**
- **Developed by:** Language Technology Research Group at the University of Helsinki
- **Model Type:** Translation (transformer-big)
- **Release**: 2022-07-28
- **License:** CC-BY-4.0
- **Language(s):**  
  - Source Language(s): dan nor swe
  - Target Language(s): heb
  - Language Pair(s): dan-heb swe-heb
  - Valid Target Language Labels: 
- **Original Model**: [opusTCv20210807_transformer-big_2022-07-28.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-heb/opusTCv20210807_transformer-big_2022-07-28.zip)
- **Resources for more information:**
  - [OPUS-MT-train GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)
  - More information about released models for this language pair: [OPUS-MT gmq-heb README](https://github.com/Helsinki-NLP/Tatoeba-Challenge/tree/master/models/gmq-heb/README.md)
  - [More information about MarianNMT models in the transformers library](https://huggingface.co/docs/transformers/model_doc/marian)
  - [Tatoeba Translation Challenge](https://github.com/Helsinki-NLP/Tatoeba-Challenge/

## Uses

This model can be used for translation and text-to-text generation.

## Risks, Limitations and Biases

**CONTENT WARNING: Readers should be aware that the model is trained on various public data sets that may contain content that is disturbing, offensive, and can propagate historical and current stereotypes.**

Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al. (2021)](https://aclanthology.org/2021.acl-long.330.pdf) and [Bender et al. (2021)](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)).

## How to Get Started With the Model

A short example code:

```python
from transformers import MarianMTModel, MarianTokenizer

src_text = [
    "Alle L.L. Zamenhofs tre bรธrn blev myrdet i holocausten.",
    "Tom visade sig vara spion."
]

model_name = "pytorch-models/opus-mt-tc-big-gmq-he"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)
translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))

for t in translated:
    print( tokenizer.decode(t, skip_special_tokens=True) )

# expected output:
#     ื›ืœ ืฉืœื•ืฉืช ื”ื™ืœื“ื™ื ืฉืœ ืืœ-ืืœ ื–ืืžื ื”ื•ืฃ ื ืจืฆื—ื• ื‘ืฉื•ืื”.
#     ืžืกืชื‘ืจ ืฉื˜ื•ื ื”ื™ื” ืžืจื’ืœ.
```

You can also use OPUS-MT models with the transformers pipelines, for example:

```python
from transformers import pipeline
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-gmq-he")
print(pipe("Alle L.L. Zamenhofs tre bรธrn blev myrdet i holocausten."))

# expected output: ื›ืœ ืฉืœื•ืฉืช ื”ื™ืœื“ื™ื ืฉืœ ืืœ-ืืœ ื–ืืžื ื”ื•ืฃ ื ืจืฆื—ื• ื‘ืฉื•ืื”.
```

## Training

- **Data**: opusTCv20210807 ([source](https://github.com/Helsinki-NLP/Tatoeba-Challenge))
- **Pre-processing**: SentencePiece (spm32k,spm32k)
- **Model Type:**  transformer-big
- **Original MarianNMT Model**: [opusTCv20210807_transformer-big_2022-07-28.zip](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-heb/opusTCv20210807_transformer-big_2022-07-28.zip)
- **Training Scripts**: [GitHub Repo](https://github.com/Helsinki-NLP/OPUS-MT-train)

## Evaluation

* test set translations: [opusTCv20210807_transformer-big_2022-07-28.test.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-heb/opusTCv20210807_transformer-big_2022-07-28.test.txt)
* test set scores: [opusTCv20210807_transformer-big_2022-07-28.eval.txt](https://object.pouta.csc.fi/Tatoeba-MT-models/gmq-heb/opusTCv20210807_transformer-big_2022-07-28.eval.txt)
* benchmark results: [benchmark_results.txt](benchmark_results.txt)
* benchmark output: [benchmark_translations.zip](benchmark_translations.zip)

| langpair | testset | chr-F | BLEU  | #sent | #words |
|----------|---------|-------|-------|-------|--------|
| dan-heb | flores101-devtest | 0.52815 | 22.9 | 1012 | 20749 |
| isl-heb | flores101-devtest | 0.42284 | 14.2 | 1012 | 20749 |
| nob-heb | flores101-devtest | 0.49492 | 19.2 | 1012 | 20749 |
| swe-heb | flores101-devtest | 0.52408 | 23.0 | 1012 | 20749 |

## Citation Information

* Publications: [OPUS-MT โ€“ Building open translation services for the World](https://aclanthology.org/2020.eamt-1.61/) and [The Tatoeba Translation Challenge โ€“ Realistic Data Sets for Low Resource and Multilingual MT](https://aclanthology.org/2020.wmt-1.139/) (Please, cite if you use this model.)

```
@inproceedings{tiedemann-thottingal-2020-opus,
    title = "{OPUS}-{MT} {--} Building open translation services for the World",
    author = {Tiedemann, J{\"o}rg  and Thottingal, Santhosh},
    booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
    month = nov,
    year = "2020",
    address = "Lisboa, Portugal",
    publisher = "European Association for Machine Translation",
    url = "https://aclanthology.org/2020.eamt-1.61",
    pages = "479--480",
}

@inproceedings{tiedemann-2020-tatoeba,
    title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
    author = {Tiedemann, J{\"o}rg},
    booktitle = "Proceedings of the Fifth Conference on Machine Translation",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2020.wmt-1.139",
    pages = "1174--1182",
}
```

## Acknowledgements

The work is supported by the [European Language Grid](https://www.european-language-grid.eu/) as [pilot project 2866](https://live.european-language-grid.eu/catalogue/#/resource/projects/2866), by the [FoTran project](https://www.helsinki.fi/en/researchgroups/natural-language-understanding-with-cross-lingual-grounding), funded by the European Research Council (ERC) under the European Unionโ€™s Horizon 2020 research and innovation programme (grant agreement No 771113), and the [MeMAD project](https://memad.eu/), funded by the European Unionโ€™s Horizon 2020 Research and Innovation Programme under grant agreement No 780069. We are also grateful for the generous computational resources and IT infrastructure provided by [CSC -- IT Center for Science](https://www.csc.fi/), Finland.

## Model conversion info

* transformers version: 4.16.2
* OPUS-MT git hash: 8b9f0b0
* port time: Sat Aug 13 00:03:50 EEST 2022
* port machine: LM0-400-22516.local