Dataset · Helsinki-NLP
English-centric OPUS-100 parallel corpus
Helsinki-NLP/opus-100
Parallel texts in 100 languages with English always serving as one side of the pair.
What's inside
OPUS-100 is an English-centric multilingual parallel text corpus. It covers 100 languages including English, with all pairs containing English as either the source or target language. Languages were selected based on the volume of parallel data in OPUS. Size ranges from 10 million to 100 million examples.
What you can do with it
- Training machine translation models from and into English
- Fine-tuning multilingual translation models
- Evaluating translation across 100 languages
- Tasks
- Translation
- Languages
- Afrikaans, Amharic, Aragonese, Arabic, Assamese, Azerbaijani, Belarusian, Bulgarian and 4 more
- Size
- 10M–100M rows
- Downloads this month
- 27.7K
- Likes
- 260
Models trained on it
- deepvk/USER-baseSentence similarity 29.4K
- mradermacher/Llama-3-Instruct-Neurona-8b-v2-i1-GGUF 2.3K
- mradermacher/Slava-Qwen3-14B-Serbian-i1-GGUF 1.3K
- mradermacher/TurkmenChat-3B-GGUF 1K
- mradermacher/Llama-3-Instruct-Neurona-8b-GGUF 840
- mradermacher/Llama-3-Neurona-8b-GGUF 788
- mradermacher/X-ALMA-13B-Group5-GGUF 788
- mradermacher/X-ALMA-13B-Pretrain-i1-GGUF 589
How to load it
from datasets import load_dataset
ds = load_dataset("Helsinki-NLP/opus-100")