Paper: https://pubs.acs.org/doi/10.1021/acscentsci.2c01123
This model predicts ionic conductivity ($\ln \sigma$), activation energy ($E_a$), and the Arrhenius prefactor ($\ln A$) for polymer electrolytes. It uses a physics-informed architecture where the output is constrained by the Arrhenius equation:

$$ \ln \sigma = \ln A - \frac{E_a}{RT} $$

## Usage

This model expects inputs processed via `MolGraphizer` and `expand_polymer_smiles`.
It requires the following features:
- **Molecular Graph**: Nodes, Edges, Connectivity.
- **Auxiliary Features**: Temperature (K), Log Molecular Weight, Molality.

To use this model for screening new candidates, use the `screen_from_hub.py` script provided in the repository.

```python
# Pseudo-code for loading
import flax.serialization
from huggingface_hub import hf_hub_download

path = hf_hub_download(repo_id="eamag/chemarr", filename="model.msgpack")
with open(path, "rb") as f:
    artifact = flax.serialization.from_bytes(dummy_state, f.read())