v2.1.2: Avoid specifying spacy as an install dependency

Specifying spaCy as a dependency causes the models to depend on spaCy as well, which causes the model artifacts to be pinned to a particular range of spaCy versions. spaCy's download already specifies which models it's compatible with, so we don't want this --- it causes spurious model repackaging and redownloading for irrelevant changes.

v0.3.0: Avoid depending on spaCy itself

Drop the direct dependency on spaCy, to avoid requirement circles.

Specifically, we're changing models to no longer specify a spaCy version as a requirement, to allow models to be forward compatible. However, the transformer models depend on this library. If it then pulls in spaCy, we end up with spaCy in the requirements again.

Relax Thinc pin to allow numpy v2

Thinc 9.0.0 is built against numpy v1, while Thinc 9.1.0 is built against numpy v2. Relax the thinc pin to allow compatibility with more recent numpy.

v2.0.0: use Curated Transformers 2.0 and discriminative learning rate schedule

✨ New features and improvements

• Rebase on Curated Transformers 2.0 (#19).
• Add the transformer_discriminative.v1 schedule (#27). This schedule uses a two different schedules:
  • transformer_schedule for transformer parameters;
  • default_schedule for other parameters.

v0.2.2: Make `DocTransformerOutput` serializable with msgpack

✨ New features and improvements

• Make DocTransformerOutput serializable with msgpack (#29).

v0.2.1: Add 'init fill-curated-transformer' CLI

✨ New features and improvements

• Add init fill-curated-transformer CLI command (#16, #22).

🔴 Bug fixes

• Correctly pass hidden_width as the embedding width in non-ALBERT models (#17).
• Add vocab_size to default transformer pipe config (#20).

v0.2.0: Support for replacing listeners

✨ New features and improvements

• Add support for replacing listeners (#7).

⚠️ Backwards incompatibilities

• Rename pipe to CuratedTransformer (#12).
• Remove quantization support until the serialization API is stabilized (#6).

Fix handling of empty docs

✨ New features and improvements

• Switch from cutlery to curated-tokenizers for word/sentencepiecing (#4).

🔴 Bug fixes

• Do not raise an exception when encountering an empty doc (#3).

Split out spaCy entry points from curated-transformers

This new package contains the entry points and spaCy/Thinc wrapping for curated-transformers that originally resided in the curated-transformers package.