The Well
A dataset, not a model: a shared collection of physics simulations for training and testing foundation models.
- Made by
- Ruben Ohana, Michael McCabe, Lucas Meyer and others (Polymathic AI)
- Field
- Simulation and fluids (a dataset, not a model)
- Learns from
- 15 TB of simulations in 16 datasets: fluids, supernovae, biology, acoustics
- Size
- Not stated
- Open weights
- Not a model, BSD-3-Clause (code)
- First public
- Nov 2024
- Source
- arxiv.org/abs/2412.00568
How it relates to Flare
Shows how simulation data gets shared across a field. Whether CERN will publish Flare's training simulations is not known.
Links
Every figure here is from the source above, checked 17 October 2026. Something changed? Email [email protected].