Trained Model Weights and Dataset-Split Metadata for "Accelerating Radiative Transfer for Planetary Atmospheres by Orders of Magnitude with a Transformer-Based Machine Learning Model"

Version 2: adds the complete raw training corpus (picaso_results_5M.h5, ~4.0 GB, HDF5): 5,000,000 atmospheric profiles with pressure grids, temperature profiles, global parameters, and the PICASO net thermal and net reflected layer fluxes used as training targets. Together with dataset_splits.json (train/validation/test indices) and normalization_metadata.json, this makes the archive fully self-contained: the validation and test inputs and targets can be reconstructed exactly, and all reported metrics can be independently reproduced.Trained model weights and dataset-provenance metadata accompanying the paper "Accelerating Radiative Transfer for Planetary Atmospheres by Orders of Magnitude with a Transformer-Based Machine Learning Model" (Malsky et al., submitted). Includes the trained transformer and LSTM model weights (best_model.pt, and the exported stand_alone_model.pt2 for the transformer), their training configurations/metadata/logs, the dataset-split indices (which PICASO-generated atmospheric profiles belong to the train/validation/test partitions), and the normalization statistics used to train the models. The data-generation pipeline (PICASO wrapper) and the emulator code are publicly available at https://github.com/imalsky/Problemulator.