Deep generative models hold great promise for inverse materials design, yet their efficiency and accuracy remain constrained by data scarcity. Here, we introduce AlloyGAN, a closed-loop framework that integrates Large Language Model (LLM)-assisted text mining with Conditional Generative Adversarial Networks (CGANs) to enhance data diversity and improve inverse design. Taking alloy discovery as a case study, AlloyGAN systematically refines material candidates through iterative screening and experimental validation. For metallic glasses, the framework predicts thermodynamic properties with discrepancies of less than 8% from experiments, demonstrating its robustness. By bridging generative AI with domain knowledge and validation workflows, AlloyGAN offers a scalable approach to accelerate the discovery of materials with tailored properties, paving the way for broader applications in materials science.

For installing, follow these intructions

conda create -n alloygan python=3.8
conda activate alloygan
pip install -r requirements.txt

The collected data in PDF format has been placed in the main directory./Collected Alloy Dataset.pdf. The complete Alloy dataset will be released later and made available for download.

Running training of GAN model and CGAN model on alloy dataset:

python main.py --model GAN --is_train True --download False   --dataroot datasets/alloys/   --dataset alloys  --epochs 100 --cuda True --batch_size 2
python main.py --model CGAN --is_train True --download False   --dataroot datasets/alloys/   --dataset alloys  --epoch 100 --cuda True --batch_size 2

Start tensorboard:

tensorboard --logdir ./logs/

Experiment validation of two metallic glass candidates, Zr_62.1Cu_31.4Al_5.1Ni_1.4 and Zr₆₂Cu_29.6Al_4.2Ag_2.3Ni_1.9

Experimental validation of metallic glass candidates generated by the AlloyGAN framework.

If you find this code useful in your research, please consider citing:

@article{hao2025inverse,
  title={Inverse Materials Design by Large Language Model-Assisted Generative Framework},
  author={Hao, Yun and Fan, Che and Ye, Beilin and Lu, Wenhao and Lu, Zhen and Zhao, Peilin and Gao, Zhifeng and Wu, Qingyao and Liu, Yanhui and Wen, Tongqi},
  journal={arXiv preprint arXiv:2502.18127},
  year={2025}
}