0.2.1

The package does not provide reliable estimates for single words or very short texts. It is meant to be used with paragraphs of text. See this issue for more details.

We added a user warning in this release if texts of 5 words or less are processed as well as a note to the README and a comment in the demo notebook.

0.2.0

Major updates and refactoring

• Upgrade to current Python and dependency versions
• Refactor LLM integration to use OpenRouter and current model lineup
• Expand synthetic data generation to 10 LLMs (up from 6)
• Simplify and improve the modelling pipeline
• Adopt uv for environment and dependency management
• Update and strengthen the test suite
• Refresh and expand documentation

Note: This release will produce different scores and may change CEFR estimations compared to v0.1.0. Overall, results should be similar while more precise and robust, but they will vary due to changes in training data and feature selection.

0.1.0

Initial release February 2025