forked from Helsinki-NLP/Tatoeba-Challenge
-
Notifications
You must be signed in to change notification settings - Fork 0
/
opus-2021-02-23.yml
38 lines (38 loc) · 980 Bytes
/
opus-2021-02-23.yml
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
release: eng-ben/opus-2021-02-23.zip
release-date: 2021-02-23
dataset-name: opus
modeltype: transformer
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
- source: spm32k
- target: spm32k
subword-models:
- source: source.spm
- target: target.spm
use-target-labels:
- >>ben<<
- >>ben_Cyrl<<
- >>ben_Deva<<
- >>ben_Gujr<<
source-languages:
- eng
target-languages:
- ben
training-data:
- eng-ben: Tatoeba-train (2660775)
- eng-ben_Cyrl: Tatoeba-train (41)
- eng-ben_Deva: Tatoeba-train (9)
- eng-ben_Gujr: Tatoeba-train (6)
validation-data:
- ben-eng: Tatoeba-dev, 2637
- total size of shuffled dev data: 2637
- devset = top 2637 lines of Tatoeba-dev.src.shuffled!
test-data:
- Tatoeba-test.eng-ben: 2500/11654
- tico19-test.eng-ben: 2100/51751
BLEU-scores:
- Tatoeba-test.eng-ben: 16.6
- tico19-test.eng-ben: 5.6
chr-F-scores:
- Tatoeba-test.eng-ben: 0.481
- tico19-test.eng-ben: 0.327