texdata/opus100-en-sl-translate-sft

Slovenian ↔ English Translation SFT (OPUS-100, instruction format)

Built by [MediaAtlas](https://mediaatlas.si/ai-training.html) — dataset engineering and LLM fine-tuning on your own data, in the EU. Pricing · All our models

120,000 bidirectional English↔Slovenian translation examples, formatted as instruction/response pairs (Alpaca style) for supervised fine-tuning (SFT).

Built by taking existing parallel pairs from [Helsinki-NLP/opus-100](https://huggingface.co/datasets/Helsinki-NLP/opus-100) (en-sl split) and wrapping them in varied translation prompts in both directions (en→sl and sl→en). No new translation was performed — the target side comes from OPUS-100.

Used to give [Sokol](https://huggingface.co/texdata) (a small general Slovenian assistant) its bidirectional translation ability.

Format

Alpaca-style, one JSON object per line:

{
  "instruction": "Translate the following English text into Slovenian:\n\nHours ago.",
  "input": "",
  "output": "Nekaj ur nazaj."
}

• —instruction — a translation prompt (varied wording) with the source text inline.
• —input — always empty.
• —output — the target translation.
• —Both directions are present (English→Slovenian and Slovenian→English), with several prompt paraphrases so the model generalises the task rather than a single template.

Provenance & domain

• —Source: OPUS-100 en-sl (a mix of subcorpora — heavily OpenSubtitles, plus others).
• —Domain: general. Note OPUS-100 includes a medical subcorpus (EMEA — drug leaflets), so a minority of pairs are medical (e.g. dosages, drug names). This is incidental to OPUS-100, not a deliberate medical set.

License

Derived from OPUS-100, which is a collection of subcorpora with varying / mixed upstream licenses. Marked license: other; verify the upstream OPUS-100 terms before commercial use or redistribution. This repo only reformats OPUS-100 pairs into instructions.

Citation

@inproceedings{zhang-etal-2020-improving,
  title = {Improving Massively Multilingual Neural Machine Translation and Zero-Shot Translation},
  author = {Zhang, Biao and Williams, Philip and Titov, Ivan and Sennrich, Rico},
  booktitle = {ACL}, year = {2020}
}