KSTAR fusion_lstm world-model weights — attribution
====================================================

weights.npz contains the KSTAR fusion_lstm neural-network ensemble weights
(βn/q95/q0/li dynamics LSTM, steady-state init net, and β-power net), converted
from the ONNX release in the NeoRL2 benchmark to a JAX-friendly format by
tools/artifacts/convert_kstar_onnx.py. The JAX reimplementation lives in
plasmax/models/world_model.py and plasmax/models/world_model_env.py.

Source
------
NeoRL2: https://github.com/polixir/NeoRL2 (neorl2/envs/data/fusion_lstm).
  Code license: Apache-2.0. Dataset/weights license: CC BY 4.0.

Cite
----
- H. Seo et al., "Feedforward beta control in the KSTAR tokamak by deep
  reinforcement learning," Nuclear Fusion 61, 106010 (2021).
- NeoRL-2: "Near Real-World Benchmarks for Offline Reinforcement Learning with
  Extended Realistic Scenarios," arXiv:2503.19267.

The weights are redistributed here under CC BY 4.0; this notice provides the
required attribution.
