Prodigy 1.13.0 is out! 🎉

Hey everyone!

We have just released Prodigy v1.13.0 which comes with a set of new recipes for LLM-assisted workflows via spacy-llm that offers multi-LLM backends as well as improved prompts for NER, textcat and spancat annotation! :sparkles:

The new spacy-llm recipes are an improved version of our existing OpenAI recipes which we'll be slowly phasing out.

There are several advantages to the new spacy-llm workflows that we'd like to highlight:

  • you can now choose from various LLM providers or even use a locally running model eliminating the need to send your data to a third party. A full list of supported models can be found on spacy-llm docs:
  • we have provided workflows for NER, textcat and spancat annotations but in custom recipes you can benefit from all spacy-llm tasks. The tasks and prompts are under active development from spaCy team so all the upgrades to spacy-llm can be expected to be directly transferable to your Prodigy workflows.
  • you are now able to improve your prompts by adding examples and label definitions for your task
  • you can also configure the cache to help reduce costs
  • all these settings (and more!) are configurable using spaCy config system which makes it easy to keep your environment reproducible and well organized:

We have updated our Large Language Models docs, where you find all details and a ton of examples.
As always, we are looking forward to your feedback :slight_smile:

To install:

pip install --upgrade prodigy -f