First check the existing inputs in
bobleesj/quantem-data.
Viewer collections use widget/; reusable images and diffraction scans stay
under haadf/ and 4dstem/. Reusing a dataset for another tutorial does not
require a new upload. The current canonical-path migration adds no new data.
Hugging Face stores scientific inputs, minimal cards and numerical metadata.
Keep notebooks, Python helpers, models and result reports in their owning source
repositories. New data contributions require permission and accurate provenance;
follow the dataset card and review the exact file list. Contribution validation
belongs in the quantem.data source checkout that provides it, not in a Python
module downloaded from the dataset. Do not assume an unpublished validator or
qualified-download API is available in an installed release.
When a notebook should load an input by name, the widget source change includes:
- helper in
src/quantem/widget/datasets.py - list it on Tutorial Datasets
- use it in the tutorial
- record its canonical folder and immutable Hugging Face revision
- link any associated data change
See Tutorial Datasets for the current pin. An explicit older revision must resolve all files within that revision, without fetching missing files from a newer snapshot.
A new widget is discuss-first. See Pull requests.
Do not commit .npy or .emd here.