AtomBench

Benchmark generative crystal reconstruction models. Score how faithfully a model rebuilds known crystals, render figures and tables, and publish results to the JARVIS-Leaderboard.

Get started →

pip install atombench
atombench mymodel.csv out/        # metrics, figures, tables

Installation

Install AtomBench and the optional submit extras.

Installation

Tutorial

Build a benchmark CSV, compute metrics, and submit — end to end.

Tutorial

API reference

The submit() and compute_metrics() Python functions.

API reference

Reproducing our benchmarks

Re-run the Snakemake pipeline on an HPC cluster.

Reproducing our benchmarks

Leaderboard

Browse the live AtomGenBench leaderboard.

https://atomgptlab.github.io/jarvis_leaderboard/Special/AtomGenBench/

GitHub

Browse the source, file issues, and contribute.

https://github.com/atomgptlab/atombench