DSPy: The framework for programming—not prompting—language models
Python
Update README.md
Omar Khattab65233b143a6b
diff --git a/README.md b/README.mdindex 8d86a8acd..d6949bb59 100644--- a/README.md+++ b/README.md@@ -15,16 +15,15 @@-----**DSPy** is a framework for developing **LM pipelines** and automatically optimizing their quality and cost. LMs are sensitive to prompts. This becomes a serious challenge in multi-stage pipelines, where different prompts (or finetunes) have to compose effectively. 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 for each step, (3) generate synthetic data to tune each step, and (4) use that data to finetune smaller LMs to cut costs.+**DSPy** is a framework for developing **LM pipelines** and automatically optimizing their quality and cost. LMs are sensitive to prompts. This becomes a serious challenge in multi-stage pipelines, where different prompts (or finetunes) have to compose effectively. 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 for each step, (3) update steps to work well together, (4) generate synthetic data to tune each step, and (5) use that data to finetune smaller LMs to cut costs.-Composing these into the right multi-step pipelines is a hard, messy process: every time you change your pipeline or your LM or your data, all prompts may need to change. To make this a more systematic process, **DSPy** separates the flow of your program (`modules`) from the parameters (LM prompts and weights) of each step. DSPy introduces new `optimizers`, which are LM-driven algorithms that can craft and tune the parameters of your pipeline, given a `metric` you want to maximize.--Using **DSPy** is an iterative process. You first define your task and the metrics you want to maximize, and prepare a few example inputs — typically without labels (or only with labels for the final outputs, if your metric requires them). Then, you build your pipeline by selecting built-in layers (`modules`) to use, giving each layer a `signature` (input/output spec), and then calling your modules freely in your Python code. Lastly, you use a DSPy `optimizer` to compile your code into high-quality instructions, automatic few-shot examples, or updated LM weights for your LM.+Composing these into the right pipelines is a hard and messy process: every time you change your pipeline, your LM, or your data, all prompts may need to change! To make this all more systematic, **DSPy** separates the flow of your program (`modules`) from the parameters (LM prompts and weights) of each step. **DSPy** introduces new `optimizers`, which are LM-driven algorithms that can craft and tune the parameters of your pipeline, given a `metric` you want to maximize.**DSPy** can routinely teach powerful models like `GPT-3.5` or `GPT-4` and local models like `T5-base` or `Llama2-13b` to be much more reliable (higher quality and/or avoiding specific failure patterns) at tasks. **DSPy** will compile the _same program_ into different few-shot prompts and/or finetunes for each LM.**What can DSPy optimizers do?** Among other things, DSPy `optimizers` can inspect your data, simulate traces through your program to collect good (or bad) examples of each step, propose new instructions for each step based on past results, finetune the weights of your LM on self-generated examples, or combine several of these to improve quality or cut cost.+**How should I use DSPy for my task?** Using DSPy is an iterative process. You first define your task and the metrics you want to maximize, and prepare a few example inputs — typically without labels (or only with labels for the final outputs, if your metric requires them). Then, you build your pipeline by selecting built-in layers (`modules`) to use, giving each layer a `signature` (input/output spec), and then calling your modules freely in your Python code. Lastly, you use a DSPy `optimizer` to compile your code into high-quality instructions, automatic few-shot examples, or updated LM weights for your LM.### Table of Contents