README.md

---
license: apache-2.0
task_categories:
- question-answering
language:
- ru
tags:
- evaluation
size_categories:
- n<1K
---


# ru-mt-bench
**ru-mt-bench** is translated version of [LMSYS MT-Bench](https://huggingface.co/spaces/lmsys/mt-bench) datasets for evaluation LLMs. The translation of the original dataset was done manually.

### Overview of the Dataset
- Original dataset: [mt_bench](https://github.com/lm-sys/FastChat/blob/main/fastchat/llm_judge/data/mt_bench/question.jsonl)
- Number of tasks in original dataset: **80**
- Number of tasks: **80**
- Format: **JSON LINES**
- Task categories: **writing, roleplay, reasoning, math, coding, extraction, stem, humanities**

# Usage
To use this dataset for evaluation follow these steps:

1. Download this [jsonl](https://huggingface.co/datasets/t-bank-ai/ru-mt-bench/blob/main/raw/question.jsonl) file.

2. Use it with original [LMSYS LLM-judge](https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge) codebase.

### Sample example

```json
{
    "question_id": 81, 
    "category": "writing", 
    "turns": ["Напишите увлекательный пост в блоге о недавней поездке на Гавайи, расскажите в нем о своих впечатлениях и достопримечательностях, которые обязательно нужно посетить.", "Перепишите свой предыдущий ответ. Начинайте каждое предложение с буквы А."]
}
```