dspy-prompting

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

DSPy Declarative Language Model Programming

DSPy：声明式大语言模型编程

Build AI systems with automatic prompt optimization from Stanford NLP.

借助斯坦福NLP团队的自动提示词优化技术构建AI系统。

When to Use

适用场景

Building complex AI systems with multiple components
Programming LMs declaratively instead of manual prompting
Optimizing prompts automatically using data-driven methods
Creating modular AI pipelines that are maintainable
Building RAG systems, agents, or classifiers with better reliability

构建包含多个组件的复杂AI系统
以声明式方式编程大语言模型，替代手动编写提示词
采用数据驱动的方法自动优化提示词
创建可维护的模块化AI流水线
构建可靠性更高的RAG系统、Agent或分类器

Quick Start

快速开始

bash

pip install dspy

bash

pip install dspy

Basic Question Answering

基础问答

python

import dspy

lm = dspy.Claude(model="claude-sonnet-4-5-20250929")
dspy.settings.configure(lm=lm)

python

import dspy

lm = dspy.Claude(model="claude-sonnet-4-5-20250929")
dspy.settings.configure(lm=lm)

Define a signature (input -> output)

class QA(dspy.Signature): """Answer questions with short factual answers.""" question = dspy.InputField() answer = dspy.OutputField(desc="often between 1 and 5 words")

qa = dspy.Predict(QA) response = qa(question="What is the capital of France?") print(response.answer) # "Paris"

undefined

class QA(dspy.Signature): """Answer questions with short factual answers.""" question = dspy.InputField() answer = dspy.OutputField(desc="often between 1 and 5 words")

qa = dspy.Predict(QA) response = qa(question="What is the capital of France?") print(response.answer) # "Paris"

undefined

Chain of Thought Reasoning

思维链推理

python

class MathProblem(dspy.Signature):
    """Solve math word problems."""
    problem = dspy.InputField()
    answer = dspy.OutputField(desc="numerical answer")

cot = dspy.ChainOfThought(MathProblem)
response = cot(problem="If John has 5 apples and gives 2 to Mary, how many does he have?")
print(response.rationale)  # Shows reasoning steps
print(response.answer)     # "3"

python

class MathProblem(dspy.Signature):
    """Solve math word problems."""
    problem = dspy.InputField()
    answer = dspy.OutputField(desc="numerical answer")

cot = dspy.ChainOfThought(MathProblem)
response = cot(problem="If John has 5 apples and gives 2 to Mary, how many does he have?")
print(response.rationale)  # Shows reasoning steps
print(response.answer)     # "3"

Core Modules

核心模块

Module	Use Case
`dspy.Predict`	Basic prediction
`dspy.ChainOfThought`	Reasoning with steps
`dspy.ReAct`	Agent-like with tools
`dspy.ProgramOfThought`	Code generation for reasoning

模块	适用场景
`dspy.Predict`	基础预测
`dspy.ChainOfThought`	分步推理
`dspy.ReAct`	类Agent工具调用
`dspy.ProgramOfThought`	面向推理的代码生成

ReAct Agent

python

from dspy.predict import ReAct

class SearchQA(dspy.Signature):
    """Answer questions using search."""
    question = dspy.InputField()
    answer = dspy.OutputField()

def search_tool(query: str) -> str:
    """Search Wikipedia."""
    return results

react = ReAct(SearchQA, tools=[search_tool])
result = react(question="When was Python created?")

python

from dspy.predict import ReAct

class SearchQA(dspy.Signature):
    """Answer questions using search."""
    question = dspy.InputField()
    answer = dspy.OutputField()

def search_tool(query: str) -> str:
    """Search Wikipedia."""
    return results

react = ReAct(SearchQA, tools=[search_tool])
result = react(question="When was Python created?")

Automatic Optimization

自动优化

BootstrapFewShot

BootstrapFewShot优化器

python

from dspy.teleprompt import BootstrapFewShot

trainset = [
    dspy.Example(question="What is 2+2?", answer="4").with_inputs("question"),
    dspy.Example(question="What is 3+5?", answer="8").with_inputs("question"),
]

def validate_answer(example, pred, trace=None):
    return example.answer == pred.answer

optimizer = BootstrapFewShot(metric=validate_answer, max_bootstrapped_demos=3)
optimized_qa = optimizer.compile(qa, trainset=trainset)

python

from dspy.teleprompt import BootstrapFewShot

trainset = [
    dspy.Example(question="What is 2+2?", answer="4").with_inputs("question"),
    dspy.Example(question="What is 3+5?", answer="8").with_inputs("question"),
]

def validate_answer(example, pred, trace=None):
    return example.answer == pred.answer

optimizer = BootstrapFewShot(metric=validate_answer, max_bootstrapped_demos=3)
optimized_qa = optimizer.compile(qa, trainset=trainset)

MIPRO Optimizer

MIPRO优化器

python

from dspy.teleprompt import MIPRO

optimizer = MIPRO(
    metric=validate_answer,
    num_candidates=10,
    init_temperature=1.0
)

optimized_cot = optimizer.compile(cot, trainset=trainset, num_trials=100)

python

from dspy.teleprompt import MIPRO

optimizer = MIPRO(
    metric=validate_answer,
    num_candidates=10,
    init_temperature=1.0
)

optimized_cot = optimizer.compile(cot, trainset=trainset, num_trials=100)

Multi-Stage Pipeline

多阶段流水线

python

class MultiHopQA(dspy.Module):
    def __init__(self):
        super().__init__()
        self.retrieve = dspy.Retrieve(k=3)
        self.generate_query = dspy.ChainOfThought("question -> search_query")
        self.generate_answer = dspy.ChainOfThought("context, question -> answer")

    def forward(self, question):
        search_query = self.generate_query(question=question).search_query
        passages = self.retrieve(search_query).passages
        context = "\n".join(passages)
        answer = self.generate_answer(context=context, question=question).answer
        return dspy.Prediction(answer=answer, context=context)

python

class MultiHopQA(dspy.Module):
    def __init__(self):
        super().__init__()
        self.retrieve = dspy.Retrieve(k=3)
        self.generate_query = dspy.ChainOfThought("question -> search_query")
        self.generate_answer = dspy.ChainOfThought("context, question -> answer")

    def forward(self, question):
        search_query = self.generate_query(question=question).search_query
        passages = self.retrieve(search_query).passages
        context = "\n".join(passages)
        answer = self.generate_answer(context=context, question=question).answer
        return dspy.Prediction(answer=answer, context=context)

Structured Output

结构化输出

python

from pydantic import BaseModel, Field

class PersonInfo(BaseModel):
    name: str = Field(description="Full name")
    age: int = Field(description="Age in years")
    occupation: str = Field(description="Current job")

class ExtractPerson(dspy.Signature):
    """Extract person information from text."""
    text = dspy.InputField()
    person: PersonInfo = dspy.OutputField()

extractor = dspy.TypedPredictor(ExtractPerson)
result = extractor(text="John Doe is a 35-year-old software engineer.")

python

from pydantic import BaseModel, Field

class PersonInfo(BaseModel):
    name: str = Field(description="Full name")
    age: int = Field(description="Age in years")
    occupation: str = Field(description="Current job")

class ExtractPerson(dspy.Signature):
    """Extract person information from text."""
    text = dspy.InputField()
    person: PersonInfo = dspy.OutputField()

extractor = dspy.TypedPredictor(ExtractPerson)
result = extractor(text="John Doe is a 35-year-old software engineer.")

LLM Providers

大语言模型提供商

python

undefined

python

undefined

Anthropic

lm = dspy.Claude(model="claude-sonnet-4-5-20250929")

OpenAI

lm = dspy.OpenAI(model="gpt-4")

Local (Ollama)

lm = dspy.OllamaLocal(model="llama3.1", base_url="http://localhost:11434")

dspy.settings.configure(lm=lm)

undefined

lm = dspy.OllamaLocal(model="llama3.1", base_url="http://localhost:11434")

dspy.settings.configure(lm=lm)

undefined

Save and Load

保存与加载

python

undefined

python

undefined

Save optimized module

optimized_qa.save("models/qa_v1.json")

Load later

loaded_qa = dspy.ChainOfThought("question -> answer") loaded_qa.load("models/qa_v1.json")

undefined

loaded_qa = dspy.ChainOfThought("question -> answer") loaded_qa.load("models/qa_v1.json")

undefined

vs Alternatives

与同类工具对比

Feature	DSPy	LangChain	Manual
Prompt Engineering	Automatic	Manual	Manual
Optimization	Data-driven	None	Trial & error
Modularity	High	Medium	Low
Learning Curve	Medium-High	Medium	Low

Choose DSPy when:

You have training data or can generate it
Need systematic prompt improvement
Building complex multi-stage systems

特性	DSPy	LangChain	手动方式
提示词工程	自动	手动	手动
优化方式	数据驱动	无	反复试错
模块化程度	高	中	低
学习曲线	中高	中	低

选择DSPy的场景：

你拥有训练数据或可以生成训练数据
需要系统性地优化提示词
构建复杂的多阶段系统

dspy-prompting

Original

Translation

DSPy Declarative Language Model Programming

DSPy：声明式大语言模型编程

When to Use

适用场景

Quick Start

快速开始

Basic Question Answering

基础问答

Define a signature (input -> output)

Define a signature (input -> output)

Chain of Thought Reasoning

思维链推理

Core Modules

核心模块

ReAct Agent

ReAct Agent

Automatic Optimization

自动优化

BootstrapFewShot

BootstrapFewShot优化器

MIPRO Optimizer

MIPRO优化器

Multi-Stage Pipeline

多阶段流水线

Structured Output

结构化输出

LLM Providers

大语言模型提供商

Anthropic

Anthropic

OpenAI

OpenAI

Local (Ollama)

Local (Ollama)

Save and Load

保存与加载

Save optimized module

Save optimized module

Load later

Load later

vs Alternatives

与同类工具对比

Resources

相关资源