Transformers and spaCy
Let's look at the new transformer-based pipelines.
We'll cover the following...
spaCy v3.0 was released with great new features and components. The most exciting new feature is undoubtedly transformer-based pipelines. The new transformer-based pipelines bring spaCy's accuracy to the state of the art. Integrating transformers into the spaCy NLP pipeline introduced one more pipeline component called Transformer. This component allows us to use all HuggingFace models with spaCy pipelines. If we recall from earlier, the spaCy NLP pipeline looks like this without transformers:
With the release of v3.0, v2 style spaCy models are still supported, and transformer-based models are introduced. A transformer-based pipeline component looks like the following:
For each supported language, transformer-based models and v2 style models are listed under the Models page of the documentation. Transformer-based models can have different size and pipeline components, just like v2 style models. Also, each model has corpus and genre information as well, ...