News and Changelog

• 1.2.0 (23/11/22)

  • Automatically attempts to extract labels from HuggingFace model config if left unspecified.

  • Added new pipeline “token_classification_transformer” for token classification, including NER, POS and other types of token predictions

    • Automatically infer NER or POS from labels and assign these to doc.ents or doc.pos if possible.

  • Renamed “classification_transformer” to “sequence_classification_transformer” to avoid confusion with new pipelines.

    • Automatically assign sequence prediction to docs.cats. Can be toggled off by setting the “assign_to_cats”: False

  • Update to documentation

• 1.1.0 (15/08/22)

  • Update following this Pull request on spacy-transformers allowing for alternate model loaders. This makes the code-base more stable towards future changes and removes existing monkeypatch.

• 1.0.0 (29/01/22)

  • Stable version of spacy-wrap launches

    • extended the test suite.

    • Added docstring to ensure credit is given to the parts of the code developed by the team at Explosion AI.

    • Minor implementations details leading the pipeline to behave more consistently with the rest of the spaCy framework, including:

      • Adding an attribute to indicate that the pipeline can’t be trained.

      • Added functionality to read and write the component to disk 💾

• 0.0.1 (29/01/22)

  • First version of spacy-wrap launches

    • Including wrappers for turning any classification transformer into a spacy transformer 🎉

    • Functionality for easily including multiple of these transformers 🌟

    • it even comes with a (minimal) documentation 📖