DSPy Integration
DSPy is a framework for programming language models created by Stanford NLP. DSPy introduces two key concepts: programming model and optimizers.
Programing model: The programming model lets you define a series of components that make a language model request. Components include input and output fields, task descriptions, and calls to a vector database like Weaviate.
Optimizers: Optimizers compile your DSPy program to tune the language model prompt and/or the weights.
DSPy and Weaviate
Weaviate is integrated with DSPy through the retriever module! You only need to connect to your Weaviate cluster through WCS or localhost, and pass in your collection:
weaviate_client = weaviate.Client("http://localhost:8080") # or pass in your WCS cluster url
retriever_module = WeaviateRM("WeaviateBlogChunk", # collection name
weaviate_client=weaviate_client)
Our Resources
Here are a few resources on using DSPy from the Weaviate team!
The resources are broken into two categories:
Hands on Learning: Content framed to build your technical understanding with end-to-end tutorials.
Read and Listen: Content designed to help develop your conceptual understanding of these technologies.
Hands on Learning
| Topic | Details |
|---|---|
| Getting Started with RAG in DSPy | Learn about the four components needed to build a DSPy program: Installation, settings, datasets with dspy.Example, LLM metrics, DSPy programming model, and optimization. Notebook and Video |
| DSPy + Weaviate for the Next Generation of LLM Apps | Build a 4-layer DSPy program for generating blog posts from queries. Notebook and Video |
| RAG with Persona | Build a compound AI system with DSPy, Cohere, and Weaviate, where you'll add a persona to the language model. Notebook and Post |
| Adding Depth to RAG Programs | Enhancing DSPy programs by integrating unique input-output examples and multiple LLMs. Notebook and Video |
| Hurricane: Writing Blog Posts with Generative Feedback Loops | Introduction to Hurricane, a web app for demonstrating generative feedback loops with blog posts. Notebook and Blog |
| Structured Outputs with DSPy | The three methods for structuring outputs in DSPy programs. Notebook and Video |
| Building RAG with Command R+ from Cohere, DSPy, and Weaviate | Overview of Command R+ with a quick RAG demo in DSPy. Notebook and Video |
| Advanced Optimizers in DSPy | Dive into optimizing DSPy programs with various techniques. Notebook |
| Llama 3 RAG Demo with DSPy Optimization, Ollama, and Weaviate | Integrating Llama3 with DSPy and optimizing prompts with MIPRO. Notebook and Video |
Read and Listen
| Topic | Details |
|---|---|
| DSPy and ColBERT with Omar Khattab! - Weaviate Podcast #85 | Omar Khattab joins the Weaviate podcast to discuss DSPy and ColBERT. Video |
| DSPy Explained | The core concepts of DSPy and walks through the introduction notebooks showing how to compile a simple retrieve-then-read RAG program and Multi-Hop RAG Program. Video |
| XMC.dspy with Karel D'Oosterlinck - Weaviate Podcast #87 | Karel D'Oosterlinck joins the Weaviate podcast to discuss IReRa (Infer-Retrieve-Rank). Video |
| Intro to DSPy: Goodbye Prompting, Hello Programming | Overview of DSPy and how it solves the fragility problem in LLM-based applications. Blog |
| Fine-Tuning Cohere’s Reranker | Generate synthetic data with DSPy to fine-tune Cohere’s reranker model. Blog |
| Your Language Model Deserves Better Prompting | Overview of the DSPy optimizers for prompt tuning. Blog |
Questions and feedback
If you have any questions or feedback, let us know in our user forum.