10_langchain_basics.py

  1"""
  210. LangChain 기초 예제
  3
  4LangChain을 사용한 LLM 애플리케이션
  5"""
  6
  7print("=" * 60)
  8print("LangChain 기초")
  9print("=" * 60)
 10
 11# ============================================
 12# 1. LangChain 구조 (코드 예시)
 13# ============================================
 14print("\n[1] LangChain 기본 구조")
 15print("-" * 40)
 16
 17langchain_basic = '''
 18from langchain_openai import ChatOpenAI
 19from langchain_core.prompts import ChatPromptTemplate
 20from langchain_core.output_parsers import StrOutputParser
 21
 22# LLM
 23llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0.7)
 24
 25# 프롬프트 템플릿
 26prompt = ChatPromptTemplate.from_template("Tell me a joke about {topic}")
 27
 28# 출력 파서
 29parser = StrOutputParser()
 30
 31# 체인 구성 (LCEL)
 32chain = prompt | llm | parser
 33
 34# 실행
 35result = chain.invoke({"topic": "programming"})
 36print(result)
 37'''
 38print(langchain_basic)
 39
 40
 41# ============================================
 42# 2. 프롬프트 템플릿
 43# ============================================
 44print("\n[2] 프롬프트 템플릿 예제")
 45print("-" * 40)
 46
 47try:
 48    from langchain_core.prompts import PromptTemplate, ChatPromptTemplate
 49
 50    # 기본 템플릿
 51    template = PromptTemplate(
 52        input_variables=["product"],
 53        template="Write a marketing slogan for {product}."
 54    )
 55    print(f"기본 템플릿: {template.format(product='smartphone')}")
 56
 57    # Chat 템플릿
 58    chat_template = ChatPromptTemplate.from_messages([
 59        ("system", "You are a helpful assistant."),
 60        ("human", "{question}")
 61    ])
 62    messages = chat_template.format_messages(question="What is Python?")
 63    print(f"\nChat 템플릿: {messages}")
 64
 65except ImportError:
 66    print("langchain 미설치 (pip install langchain langchain-core)")
 67
 68
 69# ============================================
 70# 3. Few-shot 프롬프트
 71# ============================================
 72print("\n[3] Few-shot 프롬프트")
 73print("-" * 40)
 74
 75fewshot_code = '''
 76from langchain_core.prompts import FewShotPromptTemplate, PromptTemplate
 77
 78examples = [
 79    {"word": "happy", "antonym": "sad"},
 80    {"word": "tall", "antonym": "short"},
 81]
 82
 83example_template = PromptTemplate(
 84    input_variables=["word", "antonym"],
 85    template="Word: {word}\\nAntonym: {antonym}"
 86)
 87
 88few_shot_prompt = FewShotPromptTemplate(
 89    examples=examples,
 90    example_prompt=example_template,
 91    prefix="Give the antonym of each word:",
 92    suffix="Word: {input}\\nAntonym:",
 93    input_variables=["input"]
 94)
 95
 96prompt = few_shot_prompt.format(input="big")
 97'''
 98print(fewshot_code)
 99
100
101# ============================================
102# 4. 출력 파서
103# ============================================
104print("\n[4] 출력 파서")
105print("-" * 40)
106
107parser_code = '''
108from langchain_core.output_parsers import JsonOutputParser
109from pydantic import BaseModel, Field
110
111class Person(BaseModel):
112    name: str = Field(description="Name")
113    age: int = Field(description="Age")
114
115parser = JsonOutputParser(pydantic_object=Person)
116
117# 프롬프트에 형식 지시 추가
118format_instructions = parser.get_format_instructions()
119
120prompt = ChatPromptTemplate.from_messages([
121    ("system", "Extract person info. {format_instructions}"),
122    ("human", "{text}")
123]).partial(format_instructions=format_instructions)
124
125chain = prompt | llm | parser
126result = chain.invoke({"text": "John is 25 years old"})
127# {'name': 'John', 'age': 25}
128'''
129print(parser_code)
130
131
132# ============================================
133# 5. 체인 (LCEL)
134# ============================================
135print("\n[5] LCEL 체인")
136print("-" * 40)
137
138lcel_code = '''
139from langchain_core.runnables import RunnablePassthrough, RunnableParallel
140
141# 순차 체인
142chain = prompt | llm | parser
143
144# 병렬 체인
145parallel = RunnableParallel(
146    summary=summary_chain,
147    keywords=keyword_chain
148)
149
150# 분기 체인
151chain = (
152    {"context": retriever, "question": RunnablePassthrough()}
153    | prompt
154    | llm
155    | parser
156)
157
158# 실행
159result = chain.invoke({"question": "What is AI?"})
160'''
161print(lcel_code)
162
163
164# ============================================
165# 6. RAG 체인
166# ============================================
167print("\n[6] RAG 체인")
168print("-" * 40)
169
170rag_chain_code = '''
171from langchain_community.vectorstores import Chroma
172from langchain_openai import OpenAIEmbeddings, ChatOpenAI
173from langchain_core.prompts import ChatPromptTemplate
174from langchain_core.runnables import RunnablePassthrough
175
176# 벡터 스토어
177embeddings = OpenAIEmbeddings()
178vectorstore = Chroma.from_texts(texts, embeddings)
179retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
180
181# RAG 프롬프트
182template = """Answer based on context:
183Context: {context}
184Question: {question}
185Answer:"""
186prompt = ChatPromptTemplate.from_template(template)
187
188# 문서 포맷
189def format_docs(docs):
190    return "\\n\\n".join(doc.page_content for doc in docs)
191
192# RAG 체인
193rag_chain = (
194    {"context": retriever | format_docs, "question": RunnablePassthrough()}
195    | prompt
196    | ChatOpenAI()
197    | StrOutputParser()
198)
199
200# 실행
201answer = rag_chain.invoke("What is machine learning?")
202'''
203print(rag_chain_code)
204
205
206# ============================================
207# 7. 에이전트
208# ============================================
209print("\n[7] 에이전트")
210print("-" * 40)
211
212agent_code = '''
213from langchain.agents import create_react_agent, AgentExecutor
214from langchain import hub
215from langchain.tools import tool
216
217# 커스텀 도구
218@tool
219def calculate(expression: str) -> str:
220    """Calculate a math expression."""
221    return str(eval(expression))
222
223@tool
224def search(query: str) -> str:
225    """Search the web."""
226    return f"Search results for: {query}"
227
228tools = [calculate, search]
229
230# ReAct 에이전트
231prompt = hub.pull("hwchase17/react")
232agent = create_react_agent(llm, tools, prompt)
233executor = AgentExecutor(agent=agent, tools=tools, verbose=True)
234
235# 실행
236result = executor.invoke({"input": "What is 2 + 2?"})
237'''
238print(agent_code)
239
240
241# ============================================
242# 8. 메모리
243# ============================================
244print("\n[8] 대화 메모리")
245print("-" * 40)
246
247memory_code = '''
248from langchain.memory import ConversationBufferMemory
249from langchain.chains import ConversationChain
250
251# 메모리
252memory = ConversationBufferMemory()
253
254# 대화 체인
255conversation = ConversationChain(
256    llm=llm,
257    memory=memory,
258    verbose=True
259)
260
261# 대화
262response1 = conversation.predict(input="Hi, I'm John")
263response2 = conversation.predict(input="What's my name?")
264# "Your name is John"
265
266# LCEL 메모리
267from langchain_core.runnables.history import RunnableWithMessageHistory
268from langchain_community.chat_message_histories import ChatMessageHistory
269
270store = {}
271
272def get_session_history(session_id):
273    if session_id not in store:
274        store[session_id] = ChatMessageHistory()
275    return store[session_id]
276
277chain_with_history = RunnableWithMessageHistory(
278    chain,
279    get_session_history,
280    input_messages_key="input",
281    history_messages_key="history"
282)
283'''
284print(memory_code)
285
286
287# ============================================
288# 9. 간단한 실행 예제
289# ============================================
290print("\n[9] 실행 가능한 예제")
291print("-" * 40)
292
293try:
294    from langchain_core.prompts import PromptTemplate
295
296    # 프롬프트 템플릿만 테스트
297    template = PromptTemplate.from_template(
298        "Translate '{text}' to {language}."
299    )
300
301    # 포맷팅
302    prompt = template.format(text="Hello", language="Korean")
303    print(f"생성된 프롬프트: {prompt}")
304
305    # 입력 변수
306    print(f"입력 변수: {template.input_variables}")
307
308except ImportError:
309    print("langchain-core 미설치")
310
311
312# ============================================
313# 정리
314# ============================================
315print("\n" + "=" * 60)
316print("LangChain 정리")
317print("=" * 60)
318
319summary = """
320핵심 패턴:
321    # 기본 체인
322    chain = prompt | llm | output_parser
323
324    # RAG 체인
325    rag = {"context": retriever, "question": RunnablePassthrough()} | prompt | llm
326
327    # 에이전트
328    agent = create_react_agent(llm, tools, prompt)
329    executor = AgentExecutor(agent=agent, tools=tools)
330
331주요 컴포넌트:
332    - PromptTemplate: 프롬프트 구성
333    - ChatOpenAI: LLM 래퍼
334    - OutputParser: 출력 파싱
335    - Retriever: 문서 검색
336    - Memory: 대화 히스토리
337    - Agent: 도구 사용
338"""
339print(summary)