Skip to content

armgit / dspy

Public Imported from GitHub
0

DSPy: The framework for programming—not prompting—language models

Python

Update README.md

Omar Khattabebd0eff18dce
diff --git a/README.md b/README.md
index b4c64c5b7..5a401b05c 100644
--- a/README.md
+++ b/README.md
@@ -1,8 +1,8 @@
# 🎓𝗗𝗦𝗣: The Demonstrate–Search–Predict Framework
-The **DSP** framework provides a programming abstraction for rapidly building reliable AI systems with language models (LMs) and retrieval models (RMs). In it, you write a high-level **DSP program** in a few lines of code. Your **DSP** program describes how the problem should be decomposed, and its leaves are small transformations (e.g., generate a search query to find missing information) that the LM or RM can tackle much more reliably than the overarching problem. Our [research paper](https://arxiv.org/abs/2212.14024) show this can easily lead to up to 120% gains over GPT-3.
+The **DSP** framework provides a programming abstraction for rapidly building reliable AI systems with language models (LMs) and retrieval models (RMs). In it, you write a high-level **DSP program** in a few lines of code, which describes how the problem should be decomposed. The program calls a number of small transformations (e.g., `generate a search query to find missing information`) that the LM or RM can tackle much more reliably than the overarching problem. Our [research paper](https://arxiv.org/abs/2212.14024) show this can easily lead to up to 120% gains over GPT-3.
-**DSP** takes in-context learning to the next level, especially for knowledge-intensive tasks that require search (e.g., answering questions or researching complex topics). Unlike vanilla in-context learning, you do not hard-code any few-shot prompts in **DSP**. We view "prompt engineering" the same way as hyperparameter tuning (in traditional ML). It's a final and minor step that's best done _after_ building up an effective architecture and getting its components to work together.
+**DSP** takes in-context learning to the next level, and it's been demonstrated to help considerably for knowledge-intensive tasks (i.e., those that require search, like answering questions or researching complex topics). Unlike vanilla in-context learning, you do not hard-code any few-shot prompts in **DSP**! We view "prompt engineering" the same way as hyperparameter tuning in traditional ML. It's a final and minor step that's best done _after_ building up an effective architecture and getting its components to work together.
**DSP** provides a high-level abstraction for building these architectures — with LMs and search — and its reusable primitives get the components working together on your behalf. For instance, it *annotates* few-shot demonstrations for LM calls within your arbitrary pipeline automatically. Once you're happy with things, DSP can *compile* your DSP program into a tiny LM that's much cheaper to deploy.