Skip to content

CEMP v1.1.0-paper-open

Latest

Choose a tag to compare

@github-actions github-actions released this 12 Aug 03:49

CEMP v1.1.0-paper-open

This release publishes the maintained algorithm source used by the current CEMP
pages, APIs, and task executors.

Algorithm Source

  • Restores 118 executable workflow notebooks: 12 standard MD, 8 ORCA-MD, 48
    Gaussian QC, 26 ORCA QC/manual-mode, 6 analysis/query, and 18 polymer
    generation notebooks.
  • Keeps five maintained model-inference notebooks, including an independent
    CPU ionic-liquid inference example.
  • Includes the database and model helper modules imported by these notebooks.
  • Removes old, duplicate, test, conversion, Run_All, and unreferenced
    notebooks from the public release.
  • Clears notebook outputs and execution counts and validates every code cell as
    Python source.

Configuration And Documentation

  • Adds autocompute/static/cemp_software_settings.py as the shared path
    configuration used by local and remote notebook task directories.
  • Replaces fixed production paths in ORCA-MD and prediction examples with
    documented environment variables or repository-relative model discovery.
  • Builds ionic-liquid similarity indexes directly from the loaded public
    SQLite records instead of relying on an unpublished fingerprint file.
  • Pins XGBoost 2.0.3 to match the distributed model serialization format.
  • Uses the Conda C++ runtime consistently inside Docker so mixed native
    scientific packages do not resolve against the older CentOS system runtime.
  • Adds docs/algorithms.md and a machine-readable 123-notebook allowlist in
    autocompute/public_algorithm_inventory.py.
  • Documents the official project pages and distribution requirements for
    Gaussian 16, ORCA, GROMACS, Sobtop, Multiwfn, Open Babel, Open MPI, and VMD.
    External programs are not included in CEMP. Gaussian 16 requires a separately
    obtained valid license.

Models And Data

Public database CSV files and their counts are unchanged. Model weight files
are unchanged. The model archive was repacked only to replace its duplicate
ionic-liquid inference notebook with the maintained public copy and remove
macOS-only archive metadata.

asset: release_assets/cemp_public_model_assets.tar.gz
size: 11,377,421 bytes
sha256: 8bf69f11a9c128cf788a84cc618577d8858dffc7ae8f39a40f12237adbc04062

Validation

The release gate includes:

python -m compileall -q .
python manage.py check
python manage.py test
python manage.py verify_public_release --manifest data/public_manifest.json
cd crystals/frontend && npm ci && npm run build

The public web demo remains available through Docker without installing the
optional scientific software:

docker compose up --build

The Docker path was also validated on CentOS 8 using an available host port.
That test covered data loading, demo authentication, ionic-liquid search,
polymer and ionic-liquid predictions, result export, and release verification.
An isolated SQLite validation also loaded all 231,187 paper-mode database
records and confirmed the manifest table counts, including 92,864 crystal rows
and 21,402 experimental polymer property data points.