---
license: mit
task_categories:
- other
tags:
- materials-science
- chemistry
- ai-for-science
---

# LLEMABench

LLEMABench is a benchmark suite for materials discovery tasks, introduced in the paper [LLEMA: Evolutionary Search with LLMs for Multi-Objective Materials Discovery](https://huggingface.co/papers/2510.22503). 

The benchmark consists of 14 realistic tasks across various domains, including:
- Electronics
- Energy
- Coatings
- Optics
- Aerospace

It evaluates candidates based on chemical plausibility, thermodynamic stability, and multi-objective property alignment (e.g., bandgap, formation energy, and bulk modulus).

## Links

- **GitHub Repository**: [https://github.com/scientific-discovery/LLEMA](https://github.com/scientific-discovery/LLEMA)
- **Paper**: [https://huggingface.co/papers/2510.22503](https://huggingface.co/papers/2510.22503)

## Evaluation and Usage

The official implementation provides scripts to evaluate CIF files for validity (property constraints) and thermodynamic stability.

### Validity Analysis
To evaluate CIF files against task-specific property constraints:
```bash
python calculate_validity.py --tasks "Hard, Stiff Ceramics"
```

### Stability Analysis
To analyze the thermodynamic stability of candidates from validity analysis results:
```bash
python calculate_stability.py --task <task_name>
```

## Citation

```bibtex
@inproceedings{abhyankar2026llema,
        title={LLEMA: Accelerating Materials Design via {LLM}-Guided Evolutionary Search},
        author={Abhyankar, Nikhil and Kabra, Sanchit and Desai, Saaketh and Reddy, Chandan K},
        booktitle={The Fourteenth International Conference on Learning Representations (ICLR)},
        year={2026},
        url={https://openreview.net/forum?id=TIqzhBvCNB}
}
```