Skip to content

armgit / dspy

Public Imported from GitHub
0

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

Python

Update README.md

Omar Khattab4b62c9b154d2
diff --git a/README.md b/README.md
index 18646cb60..2bf825122 100644
--- a/README.md
+++ b/README.md
@@ -15,7 +15,7 @@
----
-**DSPy is a framework for algorithmically optimizing the prompts and weights of LMs**, especially in programs that call LMs one or more times in a pipeline to solve a task. To use LMs to build a complex system without DSPy, you generally have to: (1) break the problem down into steps, (2) prompt your LM well until each step works well in isolation, (3) tweak the steps to work well together now, (4) generate synthetic data to tune each step, and (5) use that data to finetune smaller LMs to cut costs.
+**DSPy is a framework that algorithmically optimizes the prompts and weights of LMs**, especially for solving tasks with pipelines that call LMs one or more times. To use LMs to build a complex system without DSPy, you generally have to: (1) break the problem down into steps, (2) prompt your LM well until each step works well in isolation, (3) tweak the steps to work well together now, (4) generate synthetic data to tune each step, and (5) use that data to finetune smaller LMs to cut costs.
The real expressive power with LMs is in composition, but composing these operations into the right pipelines is a hard and messy process: every time you change your pipeline, your LM, or your data, all prompts (or finetuning steps) may need to change. To make this more systematic, **DSPy** does two things. First, it separates the flow of your program (`modules`) from the parameters (LM prompts and weights) of each step. Second, **DSPy** introduces new `optimizers`, which are LM-driven algorithms that can craft and tune the parameters (prompts and weights) of your pipeline, given a `metric` you want to maximize.