🩺 MedCAT v2.3.0 Release Notes

This release focuses on improving robustness, fixing key import and model loading issues, and introducing a new CLI for downloading scripts compatible with MedCAT v2.

🚀 New Feature

• MedCAT Scripts Download CLI – Added a dedicated medcat-scripts download command that ensures compatibility between script versions and MedCAT v2. These scripts (previously in working_with_cogstack) support model fine-tuning, evaluation, and related workflows. The new CLI automatically fetches the correct script version for your MedCAT installation. (#206, #210)

🐛 Bug Fixes

• Tokenizer Loading from Disk – Fixed an issue where models failed to load tokenizers correctly from disk. Previously, fallback to locally available spaCy models sometimes masked this problem or caused errors during model load. (#213)
• Legacy Conversion Imports – Made imports in the legacy model converter fully dynamic, allowing NER-only models to convert successfully even when optional extras like DeID, MetaCAT, or RelCAT are not installed. (#198, #205)
• Model Card Generation – Fixed an import error that occurred when generating model cards without MetaCAT installed. The process now skips MetaCAT-specific sections gracefully if the extra isn’t relevant. (#217)
• Embedding Linker Extras – Added missing optional extra for the embedding linker to ensure dependency correctness. (#209)

🧰 Other Improvements

• Elasticsearch Utilities – Moved Elasticsearch-related code (formerly cogstack.py in working_with_cogstack) into a separate package cogstack-es. It’s now available as three optional extras: es8, es9, and os (for OpenSearch). (#123)
• Install Target Updates – Updated install targets (in docs and derivatives) for improved consistency and clarity. (#185)
• Dependency Cleanup – Removed duplicate lines for the transformers dependency in pyproject.toml. (#204)
• Release Script Improvements – Enhanced patch release scripts for greater flexibility. (#182)
• Documentation – Updated migration guide for clarity and accuracy. (#214)

What's Changed

• build(medcat related): CU-869awn3fm Update install targets by @mart-r in #185
• feat(medcat): CU-869azdc7x: Dynamic imports for legacy conversion by @mart-r in #198
• CU-869aa22g2 Add ElasticSearch bits from working_with_cogstack by @mart-r in #123
• fix(medcat): CU-869azdc7x: Dynamic imports for legacy conversion (#198) by @alhendrickson in #205
• CU-869azvxkn: Remove duplicate lines for transformers dependency. by @mart-r in #204
• feat(medcat): CU-869azeyvz Add scripts download CLI by @mart-r in #206
• bug(medcat): CU-869b07hr0 Add optional extra for embedding linker by @mart-r in #209
• feat(medcat): CU-869b09dk4 Update scripts download by @mart-r in #210
• docs(medcat): CU-869b2fcrc Update migration guide by @mart-r in #214
• bug(medcat): CU-869b36xv7 Avoid meta cat issue when getting model card by @mart-r in #217
• bug(medcat): CU-869b2hpam Fix issue loading tokenizers off disk by @mart-r in #213
• CU-869awf45h: Update patch release script for more flexibility. by @mart-r in #182

Full Changelog: medcat/v2.2.0...medcat/v2.3.0