Physics-Informed Neural Network training with precision and optimizer switching
The project provides scripts to train PINNs while dynamically changing floating-point precision and optimizer mid‑training, facilitating systematic study of trade‑offs. It includes a unified training script, experiment launcher, and result aggregation tools, targeting researchers and engineers working on PDE solvers. By automating the switch logic, it saves time compared to manual re‑runs and offers reproducible benchmarks across conditions.
View on GitHub →michae6345-crypto/PInnns