加载中
正在获取最新内容，请稍候...
DSPy: The framework for programming—not prompting—language models
DSPy is a framework for programming—not just prompting—language models (LMs). It allows developers to compose complex LM pipelines and automatically optimize prompts, weights, and other parameters for specific tasks, dramatically improving performance and reliability compared to traditional prompting methods.
Python
Added on 2025年6月26日25,785
Stars1,982
ForksPython
LanguageProject Introduction
Summary
DSPy is a Python framework that bridges the gap between natural language prompts and robust language model applications. It introduces concepts from programming languages to allow developers to systematically build, debug, and optimize complex workflows involving LMs.
Problem Solved
Traditional prompting is brittle, hard to debug, and challenging to optimize for complex tasks. Building multi-step LM applications with raw API calls is cumbersome. DSPy addresses this by providing a programmatic way to build, test, and optimize LM-powered applications.
Core Features
Declarative LM Programming
Define complex multi-step language model pipelines declaratively, treating LMs as components rather than black boxes.
Automatic Optimization & Compilation
Automatically optimize prompts and LM weights using data and gradient-based or gradient-free techniques.
Modular Architecture
Design reusable 'modules' for common LM operations like retrieval, generation, and reasoning.
Execution Tracking
Track and debug the execution of LM programs step-by-step.
Tech Stack
Python
PyTorch
TensorFlow
Hugging Face Transformers
OpenAI API
Anthropic API
使用场景
DSPy can be applied in various scenarios where building reliable and optimizable language model pipelines is critical:
复杂推理与问答系统
Details
Build multi-step reasoning pipelines for complex question answering or analysis tasks, automatically optimizing the flow and prompts.
User Value
Achieve higher accuracy and reliability in complex reasoning tasks compared to manual prompting.
构建AI代理与自动化工作流
Details
Construct sophisticated AI agents that require chaining together multiple LM calls, tool usage, and memory.
User Value
Enable the creation of more capable and robust autonomous agents.
特定任务的提示与模型优化
Details
Optimize the performance of specific LM tasks (e.g., classification, extraction, summarization) by training the prompts and weights on relevant data.
User Value
Significantly improve task-specific performance and reduce prompting costs.
Recommended Projects
You might be interested in these projects
ThrowTheSwitchUnity
Unity is a simple C unit testing framework, designed for embedded applications but useful anywhere C is used. It focuses on simplicity and minimal resource usage.
C
45831036
datahub-projectdatahub
An open-source metadata platform to empower data discovery, data governance, and data lineage across your data and AI infrastructure.
Java
107773143
sktimesktime
sktime offers a unified framework for machine learning tasks with time series, providing a consistent interface for diverse algorithms and workflows including forecasting, classification, and regression.
Python
90101592