README.md

---
license: cc-by-nc-4.0
language:
- ru
---
# russian-instructions-10k

Russian instruction-following dataset for fine-tuning LLMs. Translated from [Alpaca Cleaned](https://github.com/gururise/AlpacaDataCleaned) with coding/math boost.

| | |
|---|---|
| **Size** | 9,975 examples |
| **Language** | Russian |
| **Format** | `instruction_ru` → `output_ru` (Alpaca-style) |
| **Translation** | Gemma 4 26B via llama.cpp API |
| **Filtering** | All coding + math examples included, remaining sampled from cleaned Alpaca |
| **License** | Same as Alpaca (CC BY-NC 4.0 / subject to OpenAI ToS) |

## Structure

Each entry has two fields:
- `instruction_ru` — Russian instruction (if original had an `input`, it's appended as `\nInput: ...`)
- `output_ru` — Russian response

## Intended Use

Recommended format is ChatML via `apply_chat_template`:

```python
messages = [
    {"role": "user", "content": instruction_ru},
    {"role": "assistant", "content": output_ru},
]
text = tokenizer.apply_chat_template(messages, tokenize=False)
```

## Sample

```json
{
  "instruction_ru": "Вам дан список вопросов, классифицируйте каждый из них по одной из следующих пяти категорий: медицина, история, наука, бизнес, искусство.\nInput: Какой стране первой удалось совершить посадку космического аппарата на Луну?",
  "output_ru": "Категория этого вопроса: история."
}
```

## Composition

| Category | Count |
|---|---|
| General | ~8,812 |
| Coding | ~560 |
| Math | ~603 |
| **Total** | **9,975** |