DSPy is a new framework for developing LLM programs. It chains LLM calls to build robust systems, where the output of one LLM call becomes the input of the next. Each LLM call acts as a function that takes text as input and produces text as output.

DSPy is a new programming model inspired by PyTorch that provides significant control over LLM programs. The Signature abstraction streamlines LLM program codebases by encapsulating prompts and structured input/output data. DSPy's compiler optimizes instructions for each LLM program component while sourcing task examples.

In this video, we will review the core concepts of DSPy and walk through the introduction notebooks that demonstrate how to compile a simple retrieve-then-read RAG program and more. I hope you enjoy it!

Notebook: https://github.com/stanfordnlp/dspy/b...