Are you using spaCy's general-purpose pipelines at work? We've heard from a lot of you that these pipelines are doing good work, but you want three things: accuracy improvements, especially on up-to-date text, a PyPi server, so you can put them in requirements, and more vectors and combinations of components.

We want to introduce a subscription service with premium models that we keep updated, exposed with an easy to use PyPi server, as a sustainable way to fund more spaCy development, especially data development.

If this sounds like something your company could use, sign up here and help make sure we're building what you really need. You'll also get early beta access, and a 20% discount for the first year once the service goes on sale.

1 MINUTE SURVEY & SIGN UP

spacy-llm v0.5: improved API & NER via Chain-Of-Thought Prompting

The new spacy-llm release lets you experiment more easily with zero-shot LLM prompting as part of your NLP pipeline. The novel Chain-of-Thought NER implementation is based on the PromptNER paper by Ashok and Lipton (2023). On an internal use-case, we found this implementation to obtain significantly better accuracy: F-score increased by up to 15%!

After a great PyData Amsterdam, we'll be at PyData NYC (Nov 1-3) with an Explosion booth, lots of swag and live demos of the upcoming Prodigy Teams. If you're attending, come and say hi! 👋

Matt will also be speaking at the Generative AI Summit in Austin, TX (Oct 26) with a talk on on how many labelled examples you need for a BERT-sized model to beat GPT-4 on predictive tasks. And at PyData Tel Aviv (Nov 14), Ines will be giving the keynote presentation.

In case you missed it...

EuroPython Keynote: LLMs from Prototype to Production

Large Language Models (LLMs) have shown some impressive capabilities and their impact is the topic of the moment. What will the future look like? Are we going to only talk to bots? Will prompting replace programming? Or are we just hyping up unreliable parrots and burning money? In her EuroPython keynote, Ines presents visions for NLP in the age of LLMs and a pragmatic, practical approach for how to use Large Language Models to ship more successful NLP projects from prototype to production today.

WATCH THE VIDEO

Curated Transformers: State-of-the-art transformers, brick by brick

curated-transformers is a new library developed by the spaCy team. It provides state-of-the-art models that are composed from a set of reusable components implemented in PyTorch, including LLMs such as Falcon, Llama and Dolly v2. Each model also features consistent type annotations for optimal developer experience.

Our annotation tool Prodigy has been getting a lot of updates recently, including new recipes for using Large Language Models to speed up training and annotation, prompt engineering via tournaments, task routing, models as annotator recipes, and much more!

As a special gift to you, we're offering 20% off personal licenses until the end of October with the discount code LLM20.

spacy-llm v0.5: improved API & NER via Chain-Of-Thought Prompting

Upcoming talks and events

In case you missed it...

EuroPython Keynote: LLMs from Prototype to Production

Curated Transformers: State-of-the-art transformers, brick by brick

New Prodigy features for LLMs in the loop