All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Removed support for Python 3.9 (EOF)
- Restore README in Python metadata
- Enable artifact attestation.
- Update Github Actions images.
- Use uv to publish (maturin deprecated itself in favor of uv).
- CLI tests.
get_condifence_allmethod to retrieve confidence thresholds for all languages.
- Sentences without alphabetical characters that cannot be identified are now labeled as
zxxinstead ofund, which is now only used for predictions under confidence thresholds. - Updated dependencies.
- Model loading in Windows.
- Add new languages documentation.
- Model singleton shared between Identifier instances for Python API.
- Collapse macrolangs in top-k.
- Python API methods refactor.
- Updated dependencies.
- Re-compile model if heliport-model subpackage changed.
- Optional check all langs have threshold before loading.
- Fix confidence file parsing on Windows.
- Other small changes.
- Download feature.
- Python 3.8 support.
- More Python identifier functions:
identify_with_score,identify_topk_with_score,par_identify,par_identify_with_score. - Batched and multithreaded Python prediction.
- Rust identifier functions won't return
Optionanymore, as there was no case whereNonewas used. - Included
Cargo.lock. - Update Github Actions build container to Python 3.12.
- Update Github Actions to MacOS 13.
- Update to Maturin 1.8.
- Small documentation issues.
sdistproject layout withoutheliport.data
- Model creation command.
- More verbosity during identification.
- Include binarized model in the wheel.
- Binarize model during compilation.
- Separate languagemodel stuff in a subpackage.
- Disable download feature by default.
- Use Rustls for download instead of OpenSSL (less hassle when building from source).
- Parallelize model binarization.
- Update Python bindings to PyO3 0.23.
- Fix compilation without python feature.
- Min Dong missing as CJK language.
- Removed old entrypoints.
- Faster
score\_langtriggering autovectorization. - Multithreaded identification.
- Load only relevant languages feature.
- More verbosity when binarizing models
- Support i/o files other than stdin/stdout.
- Rank top k languages (still not available in the CLI).
- Alias
detectcomand foridentify. - More verbosity when binarizing models
- MSRV to 1.71.
- Bring back HeLI models (use 2.0).
- Refactored the CLI into subcommands
- Renamed
compilecommand tobinarize - Better error propagation with Anyhow.
- Separate CLI, download and python into crate features.
- Renamed Model to ModelNgram and Models to Model.
- Update tests.