This repository is provided in a tentative form and will be updated soon.
This repository contains the author’s implementation of the paper “Bridging Text and Crystal Structures: Literature-Driven Contrastive Learning for Materials Science.” It includes all training and evaluation scripts for CLaSP (Contrastive Language-Structure Pre-training), proposed in the publication.
Please refer to docker/Dockerfile for the complete software environment. All required Python versions and libraries are defined within the Dockerfile.
# From repository root:
# 1. Build Docker image
docker build -t clasp:latest -f docker/Dockerfile .
# 2. Start container (with GPU support if available)
docker run --gpus all \
-v $(pwd):/workspace/clasp \
-w /workspace/clasp \
-it clasp:latest bashto be updated
If you use this code, please cite the paper using the following BibTeX entry:
@misc{suzuki2025contrastivelanguagestructurepretrainingdriven,
title={Contrastive Language-Structure Pre-training Driven by Materials Science Literature},
author={Yuta Suzuki and Tatsunori Taniai and Ryo Igarashi and Kotaro Saito and Naoya Chiba and Yoshitaka Ushiku and Kanta Ono},
year={2025},
eprint={2501.12919},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2501.12919},
}- Publish training and evaluation scripts
- Release pre-trained model weights
- Release dataset
- Add examples