smalleval/mmlu-nano

# SmallEval: Browser-Friendly LLM Evaluation Datasets 🚀

[](https://cloudcode.ai)

SmallEval is a curated collection of lightweight evaluation datasets specifically designed for testing Large Language Models (LLMs) in browser environments. Each dataset is carefully subsampled to maintain a small footprint while preserving the evaluation quality.

## 🎯 Purpose

The primary goal of SmallEval is to enable efficient evaluation of LLMs directly in web browsers. Traditional evaluation datasets are often too large for browser-based applications, making it challenging to assess model performance in client-side environments. SmallEval addresses this by providing:

- Compact dataset sizes (250 samples per subset)
- Carefully selected samples from established benchmarks
- Browser-friendly JSONL format
- Consistent evaluation metrics across different domains

## 📊 Available Datasets

Each dataset is a subset of the original LightEval collection, containing 250 randomly sampled examples:

### MMLU (Massive Multitask Language Understanding)
- `mmlu_high_school_mathematics.jsonl`
- `mmlu_high_school_physics.jsonl`
- `mmlu_high_school_biology.jsonl`
- `mmlu_high_school_chemistry.jsonl`
- `mmlu_high_school_computer_science.jsonl`
- `mmlu_high_school_psychology.jsonl`
- `mmlu_high_school_us_history.jsonl`
- `mmlu_high_school_world_history.jsonl`

## 📥 Usage
Checkout our Github Repo: https://github.com/Cloud-Code-AI/smalleval

## 🤝 Contributing

We welcome contributions! If you'd like to add new subsets or improve existing ones, please:
1. Fork the repository
2. Create your feature branch
3. Submit a pull request

## 📜 License

These datasets are derived from the original [LightEval](https://huggingface.co/lighteval) collection and maintain their original licenses.

## 🔗 Links

- [Cloud Code AI](https://cloudcode.ai)
- [Original LightEval Datasets](https://huggingface.co/lighteval)