| --- |
| license: other |
| license_name: cc-by-sa-4.0-and-odc-by-see-sources |
| language: |
| - sl |
| tags: |
| - slovenian |
| - pretraining |
| - continued-pretraining |
| - text |
| - research |
| pretty_name: Slovenian Continued-Pretraining Corpus |
| task_categories: |
| - text-generation |
| size_categories: |
| - 100K<n<1M |
| --- |
| |
| # Slovenian Continued-Pretraining Corpus — slo-pretrain |
|
|
| > Built by **[MediaAtlas](https://mediaatlas.si/ai-training.html)** — dataset engineering and LLM fine-tuning on your own data, in the EU. [Pricing](https://mediaatlas.si/ai-training-pricing.html) · [All our models](https://huggingface.co/texdata) |
|
|
| A cleaned **monolingual Slovenian text** corpus used for **continued pre-training (CPT)** |
| of the Qwen3.6 base before Slovenian SFT (part of the Vran / Slovenian-model line). |
|
|
| > **General Slovenian, not medical.** This is a general-domain Slovenian text corpus for |
| > language adaptation. For the medical instruction data see |
| > [`texdata/med-slo-sft`](https://huggingface.co/datasets/texdata/med-slo-sft). |
| > Research / non-commercial use. |
|
|
| ## Contents |
|
|
| | file | rows | format | |
| |---|---|---| |
| | `data/train.jsonl` | 160,539 | `{"text": "..."}` plain text documents | |
|
|
| ~424 MB. Lightly cleaned (whitespace/unicode), one document per line. |
|
|
| ## Sources |
|
|
| - **Slovenian Wikipedia** — CC BY-SA 4.0 (attribution + ShareAlike) |
| - **FineWeb2 (sl)** — ODC-BY 1.0 |
|
|
| Attribute both. Comply with each source's terms before any use; **research and |
| non-commercial** only. |
|
|
| ## Usage |
|
|
| Standard next-token continued-pretraining on the `text` field. Used here to adapt a |
| Qwen3.6 base to Slovenian fluency/knowledge prior to domain SFT. |
|
|
| ## License |
|
|
| `license: other` — the corpus inherits the **most restrictive** upstream license |
| (CC BY-SA 4.0 from Wikipedia → ShareAlike applies). Research and non-commercial use. |
|
|
