Skip to content

Changelog

All notable releases of lumina-sdk are documented here.

The format follows Keep a Changelog, and the project adheres to PEP 440.

[v0.1.0rc1] — 2026-04-30

First public release candidate.

Added

  • Datasets — OPFDataset (in-memory), OPFOnDiskDataset (SQLite/RocksDB), OPFShardedIterableDataset, plus a to_float32 PyG transform for fp64 → fp32 casting.
  • Models — Heterogeneous GNN architectures: OPFHeteroGNN (with sage / gcn / gin / gat backends), RGAT, HEAT, HGT; homogeneous variants under OPFHomoGNN.
  • Losses — OPFLossManager with MSE / RMSE / MAE / MAPE / SmoothL1; PhysicsInformedLoss skeleton with quadratic / absolute / log-barrier penalties.
  • Trainer — Plain PyTorch DDP via MPI; gradient clipping; non-finite loss handling; W&B integration; sample-based scheduling.
  • Evaluator — ACOPFConstraintEvaluator for voltage / generation / power-balance / thermal-limit violations; Modeler for checkpoint loading and end-to-end prediction.
  • Examples — train_opf_simple.py (single-process smoke test), train_opf_ddp.py (multi-GPU DDP), evaluate_opf_constraint.py (constraint metrics).
  • Backends — Tested on CPU and NVIDIA CUDA; HPC job templates for Polaris (PBS / MPICH) and Perlmutter (SLURM).
  • Docs — Material for MkDocs site at https://argonne-gridfm.github.io/lumina-sdk/ with quickstart, training tutorial, evaluation, multi-case, HPC, and 6 runnable Jupyter notebooks.