Milyte

Milyte is a multiscale, physics-native deep learning framework for battery electrolyte property prediction and formulation design — connecting molecular solvation structure, mesoscopic aggregation, bulk transport properties, and device-level performance within a single differentiable model. It predicts ionic conductivity, viscosity, desolvation free energy, and dual-temperature DC internal resistance (DCR), and supports gradient-based inverse design (GALO) of electrolyte formulations for a target property.

This repository hosts the pretrained weights and datasets (RDF pretraining data, downstream experimental datasets, and finetuned checkpoints) used in our manuscript.

Code, training, and inference instructions are available on GitHub:

💻 github.com/xiaomi-research/milyte

Citation

If you use these weights or datasets in your research, please cite our paper (citation details to be added upon publication).

License

Released under CC BY-ND 4.0.