Skip to content

Release artifacts

HemiSpec is published as software artifacts, not only as a GitHub source repository. The primary v0.1.0 public artifact is the hemispec-toolkit Python package on PyPI, because model-enabled use is best handled inside a Python/PyTorch environment. GitHub Release artifacts are retained as archived, offline, or Windows-folder fallback artifacts at https://github.com/mqqq333/HemiSpec/releases/tag/v0.1.0.

v0.1.0 public artifacts

python -m pip install hemispec-toolkit

hemispec_toolkit-0.1.0-py3-none-any.whl          Python wheel
hemispec_toolkit-0.1.0.tar.gz                    source distribution
HemiSpec-CLI-v0.1.0-win64.exe                    Windows CLI executable
HemiSpec-GUI-v0.1.0-win64.zip                    compiled GUI folder distribution
HemiSpec-v0.1.0-SHA256SUMS.txt                   checksums
HemiSpec-v0.1.0-RELEASE_ARTIFACTS.txt            verification and artifact manifest
HemiSpec-Assets-<version>.zip                    optional offline/custom asset bundle for non-default assets

The same public Python API powers the PyPI-installed CLI, PyPI-installed GUI launcher, and any compiled fallback app so that examples and validation behavior remain reproducible.

ANS/RNS metric usage should keep citation boundaries clear: the original ANS/RNS and cross-hemispheric DGN framework comes from Wang et al. 2024, Patterns; HemiSpec packages and extends that workflow for the current software release.

Build commands

Build the lightweight package and, unless skipped, compiled Windows artifacts with:

powershell -ExecutionPolicy Bypass -File scripts\build_release.ps1 -Python "py -3.12"

Useful switches:

powershell -ExecutionPolicy Bypass -File scripts\build_release.ps1 -Python "py -3.12" -SkipExe
powershell -ExecutionPolicy Bypass -File scripts\build_release.ps1 -Python "py -3.12" -SkipGuiSmoke

-SkipGuiSmoke should only be used in a headless environment where GUI launch checks are impossible.

Release acceptance gates

A release is not considered public-ready until all relevant artifacts are built and checked together:

  • python -m build --wheel --sdist creates lightweight hemispec-toolkit artifacts.
  • python -m twine check dist/hemispec_toolkit-*.whl dist/hemispec_toolkit-*.tar.gz validates package metadata before upload.
  • python -m twine upload dist/hemispec_toolkit-*.whl dist/hemispec_toolkit-*.tar.gz publishes the package to PyPI.
  • pip install dist/*.whl or pip install hemispec-toolkit exposes hemispec and hemispec-gui entry points in the active Python/PyTorch environment.
  • hemispec quickstart --out-dir <tmpdir> runs the built-in public-safe synthetic smoke test without a source checkout.
  • hemispec --help and documented subcommands run in a clean environment.
  • hemispec-gui starts the compact standard-workflow GUI.
  • The Windows app is built as a folder distribution and includes only approved runtime files.
  • Any additional/offline DGN model, atlas, classifier, or example data bundle has a manifest, checksums, license/provenance notes, and a compatibility version.
  • Public artifacts and docs pass leak checks for private paths, keys, subject data, model payloads, and unpublished result claims.

Source versus assets

The source repository contains code, documentation, tests, examples, and the approved reusable model bundles under Git LFS. Atlas NIfTI files, non-public neuroimaging derivatives, and any additional/custom model bundles should be released separately with checksums, licenses, and model cards unless explicitly approved for the repository.

Desktop fallback variants

  • Lightweight app: compact GUI plus CLI/API utilities for fallback/demo use; released models are resolved from Git LFS, cache download, or user-configured paths.
  • Model-enabled app: compact GUI plus approved model/atlas assets and PyTorch runtime, packaged as a larger folder distribution or paired with an offline asset bundle when PyPI/conda installation is not practical.

Both variants should use the public hemispec package internally; they are not the preferred path for users who can manage a Python environment.

Post-release verification

The v0.1.0 release was re-downloaded from GitHub on 2026-06-29. SHA256 checksums matched, the downloaded Windows CLI printed --help, and the downloaded wheel ran the public-safe synthetic quickstart in a fresh local verification workspace. See v0.1.0 release verification.