README.md

---
pretty_name: Alloy-Bench
language:
- ru
license: cc-by-nc-sa-4.0
task_categories:
- text-generation
tags:
- metallurgy
- mining
size_categories:
- 1K<n<10K
---

## Alloy-Bench

**Alloy-Bench** is a Russian-language **multiple-choice benchmark** for evaluating large language models in the domain of metallurgy and mining engineering.  
It is designed in the spirit of MMLU-style exams and focuses on professional knowledge rather than general trivia.   
This is the validation split - publicly available for open evaluation and comparison.   

### Version      
v1.1 fixed


### 📊 Dataset Statistics

- **Total questions:** 1,120 questions  
- **Language:** Russian  
- **Format:** Parquet

![Subdomain distribution](images/area_distribution_val.png)

---

### 📋 Data Format

The dataset is stored in a tabular format (Parquet). Each row corresponds to a **single multiple-choice question** and includes:

- **question** – the question text in Russian;
- **options** – a list of possible answers (3–5 choices);
- **correct_answer** – the correct option;
- **knowledge_area** – subdomain (e.g. *Металлургия редких металлов*, *Металлургия тяжелых цветных металлов*, *Химическая инженерия*).


---


### 📚 Citation

If you use Alloy-Bench in academic work or reports, please cite it as:

```bibtex
@misc{alloybench2025,
  title        = {Alloy-Bench: Russian Benchmark for Metallurgy and Mining Question Answering},
  author       = {nn-tech},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/nn-tech/Alloy-Bench}},
  note         = {Multiple-choice evaluation benchmark for domain LLMs}
}
```

---