Spaces:
Running
title: Princeton MagNet
emoji: 🧲
colorFrom: yellow
colorTo: red
sdk: static
app_file: index.html
license: mit
short_description: Magnetic-material data, models, and browser-side prediction
Princeton MagNet
This Space hosts the MagNet website: a public catalog of magnetic-material measurements, training-ready datasets, models, and documentation. Its steady-state iGSE predictor runs in your browser; the ten original steady-state B-H Transformer models also run in the browser as an experimental preview. One Predict form shows both volumetric-loss estimates and plots the Transformer's predicted B–H loop. The neural conversion was checked numerically against the preserved PyTorch weights, but its independent predictive accuracy has not been certified. No server-side prediction API is provided.
The full downloadable data are published separately in the MagNet dataset repository under CC BY 4.0. This Space contains the website, browser runtime, ten small converted model packages, and their operating-domain hulls, not the measurement archive. The website, legacy model source, and bundled ONNX Runtime Web are MIT licensed; the dataset license is separate.
The ten converted models, original PyTorch checkpoints, normalizers, operating- domain hulls, and a standalone inference example are downloadable from the MagNet Transformers model repository.
Only ten original materials have neural weights; the five newer steady-state materials and all transient materials do not yet have online neural models. Research context and the original project are available from Princeton MagNet and the PrincetonUniversity/magnet repository.