---
license: mit
tags:
  - pytorch-lightning
  - neural-pde-solver
  - operator-splitting
---

# DISCO model checkpoints

Pretrained checkpoints for the ICML 2026 paper *Test-time Generalization for
Physics through Neural Operator Splitting* (Serrano, Han, Oyallon, Ho, Morel).

| Equation | Checkpoint |
|---|---|
| Advection-diffusion (1D) | `advection-diffusion/last.ckpt` |
| Combined equation (1D) | `combined-equation/last.ckpt` |
| Reaction-diffusion (Gray-Scott, 2D) | `reaction-diffusion/last.ckpt` |
| Navier-Stokes / Euler (2D) | `navier-stokes/best-checkpoint.ckpt` |

Load with PyTorch Lightning:

```python
from train.train_generic import DISCOLitModule
model = DISCOLitModule.load_from_checkpoint("last.ckpt")
```

Code: https://github.com/LouisSerrano/disco-ttg