CrystaLLM-pi (AI-driven Prediction of Crystal Structure from Composition and Experimental Data)

CrystaLLM-pi is an open-source, transformer-based tool for generating crystalline materials directly in CIF format using large language model techniques. It enables rapid exploration of crystal structure space and supports both unconditional generation and property-conditioned design, allowing users to target materials with specific characteristics such as band gap, density, photovoltaic efficiency, or XRD patterns. Designed for materials scientists, it provides a fast, data-driven complement to traditional simulation methods, helping to generate candidate structures for further study or interpret experimental data. CrystaLLM-pi is available as source code, a prebuilt container for local deployment, and a web application for interactive use.

Guidance

Guidance for CrystaLLM-pi web application

CrystaLLM-pi github readme

You can see the resources that are part of this resource theme below.

Sort by

Showing 3 of 3 resources

Services

PXRD-conditioned CrystaLLM-pi Web Application (AI-driven Prediction of Crystal Structure)

The CrystaLLM-pi web application provides an accessible, browser-based interface for generating crystalline structures using state-of-the-art generative AI. Users can quickly create candidate materials either from chemical composition alone or by combining composition with experimental X-ray diffraction (XRD) data. Designed for ease of use, the webapp enables interactive exploration of crystal structures without requiring local installation or machine learning expertise. Generated structures are returned in standard CIF format, making them ready for visualisation or downstream simulation, and supporting rapid integration into experimental and computational materials workflows.

Tools

CrystaLLM-pi (AI-driven Prediction of Crystal Structure) Github Repository

Github repository for CrystaLLM-pi code - a generative model for producing plausible structures for crystalline materials. It can be run in several ways: de novo to generate a plausible material with no constraints; to propose a crystal structure for a given chemical composition; or to produce a structure that is consistent with a supplied X ray diffraction (XRD) pattern. The GitHub repository provides the full source code for users who wish to install and run it in their own environment.

Docker Container of CrystaLLM-pi (AI-driven Prediction of Crystal Structure)

Docker container for CrystaLLM-pi - a generative model for producing plausible structures for crystalline materials. It can be run in several ways: de novo to generate a plausible material with no constraints; to propose a crystal structure for a given chemical composition; or to produce a structure that is consistent with a supplied X ray diffraction (XRD) pattern. The Docker container provides a ready made environment for running CrystaLLM-pi locally. It bundles all required dependencies and configuration so users can run the model without setting up the software stack themselves, making it easier to test the tool, reproduce results, or integrate it into existing workflows.