gidind-ai publication checkpoint
This repository contains the PyTorch checkpoint used by gidind-ai to
propose crystallographic unit-cell candidates from grazing-incidence X-ray
diffraction (GIXD/GIWAXS) peak lists. The accompanying gidind-ai software
performs candidate generation followed by gradient/Laplace refinement; it does
not use TensorFlow or MCMC refinement.
Checkpoint
| Field | Value |
|---|---|
| File | default_model.ckpt |
| Repository revision | 374aed64cf921eb47e3e03452c4337e0e636ff93 |
| SHA-256 | b7efe2de22fdac9ece25b954634ffbc0a6df5570b01b80a2212a80cdae34dc2a |
| Framework | PyTorch / PyTorch Lightning checkpoint |
The filename and revision above are the stable reviewer-facing release. Pin this commit hash in reproducibility work rather than relying on the repository default branch.
Use with gidind-ai
Install the accompanying source package, including the optional Hugging Face download support, then run an approved peak-list example:
python -m pip install '.[models]'
gidind-ai infer \
--input examples/real_cases/pq.csv \
--output-dir reviewer_outputs/pq \
--device cpu
The command downloads this exact checkpoint automatically. To state the source explicitly, use:
gidind-ai infer \
--input examples/real_cases/pq.csv \
--output-dir reviewer_outputs/pq \
--model-repo epfeiler/gidind-ai \
--revision 374aed64cf921eb47e3e03452c4337e0e636ff93 \
--device cpu
gidind-ai writes a JSON result with checkpoint provenance, initial and
refined cells, Laplace uncertainties, convergence state, and matching metrics.
The associated repository also provides four paper-case wrappers and a small,
fixed-seed CPU benchmark.
Intended use and limitations
This checkpoint is intended for research use with the supplied gidind-ai workflow and peak-list format. It provides candidate unit cells, not a final structural determination. Results depend on peak extraction, calibration, orientation hypotheses, and the physical appropriateness of the experimental sample; they should be checked by a domain expert. It is not a substitute for crystallographic validation.
The checkpoint is a PyTorch/Lightning serialization file. Only load it from this trusted, pinned release.
License and citation
The model weights are released under CC BY-SA 4.0. Reuse requires attribution and distribution of adaptations under the same license. Please cite the associated manuscript and the gidind-ai software release when they become available.