MEIDNet 2.3.1

• meidnet score: targets.csv takes per property a point target <p>_target and/or a window <p>_min / <p>_max. A bound such as "formation energy at most 1.0 eV/atom" is a window with only <p>_max; a value inside the window is a success.
• The report table is plain ASCII and says per property whether the window or the tolerance decided.
• Run bundles from MEIDNet Matter 0.2.0 use this layout.

2.3.0 introduced meidnet score (LeMat-GenBench metric families + conditional extension) and the Learn · Benchmark · Develop · Matter navigation.

pip install meidnet==2.3.1

MEIDNet 2.3.0

• meidnet score DIR: generation-quality report for a folder of CIF files from any model, in the metric families of LeMat-GenBench (validity, uniqueness, novelty, diversity, distribution, stability with --mlip, SUN) plus MEIDNet's conditional extension from targets.csv (target success, target error, multi-property success, constraint success, conditional diversity, target coverage, interpolation share). The compatibility page defines every metric and states where this implementation differs.
• Prism navigation simplified to Learn · Benchmark · Develop · Matter ↗; new ecosystem and develop pages.
• Companion application: MEIDNet Matter (https://babu09-meidnet-matter.hf.space/).

pip install meidnet==2.3.0

MEIDNet 2.2.0

First release on PyPI: pip install meidnet.

• MEIDNet Prism platform: landing page, Studio, Learn, datasets and benchmarks (https://babu09-meidnet.hf.space/)
• Benchmark protocol perov5-v1.1 with seven-seed alignment results and analysis pages
• Ask PRISM help panel, live 3D tour, documentation redesign
• Studio: /health route; libraries served locally when the docs are present

Install: pip install meidnet (add [stability] for MACE screening). Try: meidnet demo, meidnet studio.
Paper: https://doi.org/10.1038/s41524-026-02153-3

MEIDNet 2.1 - MEIDNet Prism

MEIDNet 2.1 — MEIDNet Prism

The interactive platform around the MEIDNet framework, live at https://babu09-meidnet.hf.space/.

New

• Landing page: what MEIDNet does in one minute (animated tour, also the attached meidnet_tour.webm), Explore / Build / Benchmark.
• Studio at /studio/: upload your table, train in the browser, rules, targets, latent search, candidates in 3D, Researcher mode with the configuration as text.
• Explore: datasets (Perov-5, MP-20, Carbon-24) and a capabilities page generated from the code.
• MEIDNet Benchmarks: results per dataset with an evidence ladder; community submissions through GitHub (scripts/benchmarks.py).
• Pretrained Perov-5 checkpoints on the Hub: https://huggingface.co/Babu09/MEIDNet (attached here as a zip too).

Paper: A. Babu, R. A. Gouvêa, P. Vandergheynst, G.-M. Rignanese, npj Computational Materials (2026), doi:10.1038/s41524-026-02153-3.

Install: pip install git+https://github.com/ABnano/MEIDNet.git@v2.1.0 then meidnet demo.