I’ve recently jumped in to understanding DSPy. Their tag line is “Programming—not prompting—Foundation Models”. My first impression is that it is a solid framework for coordinating LLM workflows in a completely new way. And the tag line, so far, “feels” very right!

These are some resources that helped me get an initial impression. Many of the docs are very clear and easy to follow. I’ll try to keep this list updated as I go:

DSPy Links

DSPy Github Repo - the readme is really great, don’t skip reading it :)

Intro Notebook - solid intro to concepts

DSPy - Compiling Declarative Language Model Calls into Self-Improving Pipelines (Oct 2023)

DSPy Assertions - Computational Constraints for Self-Refining Language Model Pipelines (Dec 2023)

In-Context Learning for Extreme Multi-Label Classification - really cool application of DSPy on extreme multi-label classification with thousands of classes, authors achieved SOTA results on several benchmarks…this was the result that got me initially interested in DSPy (Jan 2024)

Compiling for tricky tasks - very interesting notebook

DSPy Guides - looking forward to getting to these at some point

Announcing DSPy - tweet

DSPy paper reviewer comments - interesting read and discussion

Redefining Language Model interactions through DSPy - blog gives an overview

Introducing Assertions in DSPy - blog gives intro to concepts covered in the assertions paper

DSPy Docs - @lateinteraction provided the link to these docs, they’re also really good and thorough!

DSPy walkthrough by Kanav - initial insights from reading the paper and playing around with the tutorials

DSPy one-page cheat sheet - Dataloaders, Evaluation, Optimizers, Assertions and more

DSPy Videos

DSPy - solid intro video

DSPy Explained - by Connor Shorten, recommended

Omar talking about DSPy on ML Ops Podcast - ML Ops is great, Austin conference was 🔥

DSPy and ColBERT with Omar Khattab - Weaviate Podcast

Karel D’Oosterlinck discussing Extreme Multi-Label Classification - Weaviate Podcast