Getting Started
Download training data, train a surrogate, and submit for automatic holdout scoring on standardized physics problems. For background on the platform, see the About page.
Quick start guide
The full loop from data download to a row on the problem page.
Browse problems
Pick a live problem from the showcase. Each card has a short description, a link to the problem page, and (on the problem page) a download for the training data.
View problemsDownload training data
Each problem ships a public train split as a compressed archive (.tar.zst). You get every training case in Rescale AI Physics layout: meshes, per-case YAML, and a case index CSV. Holdout meshes are not included.
Training dataTrain your model
Train on the public split with Rescale AI Physics or your own stack. Use any architecture you like, or browse prior submissions on the problem page for ideas on what others have tried.
Package and submit
Pack a BYOM archive with submission.json and a plugin/ folder, then upload at Submit. See the submission guide for the exact layout and plugin contract.
Submission guideTrack and compare
Follow your eval job on the track page. Published runs appear on each problem page with holdout scores and per-case breakdowns.
View submissionsNext steps
- Training data formats
What each archive contains, file types, and units
- Architecture guide
Pros, limitations, and problem fit for each surrogate family
- Evaluation protocol
Reference hardware, scoring rules, and available metrics
Proposing a new problem
Have simulation data you want on the platform? We work with you on train and test splits, keep holdout cases private on our side, and publish a fair public training pack. Contact us to start the conversation.
Need help?
If you have questions or run into issues, check or contact: