LangChain Tutorial - The SECRET to Building Viral LLM Apps

Learn the secret techniques top AI developers use to easily manage LLMs, optimize prompts, chain models, and create conversational apps with 10x better results.

What is 🦜🔗LangChain?

!pip install langchain openai huggingface_hub wikipedia chromadb faiss-cpu tiktoken -q

Downloading Wikipedia Page

from pathlib import Path import requests response = requests.get( "https://en.wikipedia.org/w/api.php", params={ "action": "query", "format": "json", "titles": "Gregory_Piatetsky-Shapiro", "prop": "extracts", # 'exintro': True, "explaintext": True, }, ).json() page = next(iter(response["query"]["pages"].values())) italy_text = page["extract"] data_path = Path("data") if not data_path.exists(): Path.mkdir(data_path) with open("data/Gregory.txt", "w") as fp: fp.write(italy_text)

LLMs

Using OpenAI GPT-3.5-turbo LLM to answer the general question.

from langchain.llms import OpenAI llm = OpenAI(temperature=0.9) text = "What do you know about KDnuggets?" print(llm(text))

You can also use other LLMs like HuggingFace LLaMA-2-7B-32K

from langchain import HuggingFaceHub llm = HuggingFaceHub(repo_id="togethercomputer/LLaMA-2-7B-32K", model_kwargs={"temperature":0, "max_length":64}) print(llm("How old is KDnuggets?"))

Prompt Templates

LangChain enables robust prompt engineering and optimization through its flexible, prompt template system.

from langchain import PromptTemplate template = """Question: {question} Make the answer more engaging by incorporating puns. Answer: """ prompt = PromptTemplate.from_template(template) llm(prompt.format(question="Could you provide some information on the impact of global warming?"))

Chains

LangChain allows chaining LLMs together with prompt templates to create complex multi-step workflows.

from langchain import LLMChain llm_chain = LLMChain(prompt=prompt, llm=llm) question = "Could you provide some information on the impact of global warming?" llm_chain.run(question)

Agents & Tools

LangChain Agents interface with external tools and services, using the LLMs to determine optimal sequences of actions to accomplish tasks like search, calculations, and data lookups.

from langchain.agents import load_tools from langchain.agents import initialize_agent tools = load_tools(["wikipedia", "llm-math"], llm=llm) agent = initialize_agent(tools, llm, agent="zero-shot-react-description", verbose=True) agent.run("Can you tell me the distance between Earth and the moon? And could you please convert it into miles? Thank you.")

Memory

LangChain provides simplified state management across chained LLMs and agent calls through a consistent memory interface, enabling smooth workflows.

from langchain.chains import ConversationChain from langchain.memory import ConversationBufferMemory conversation = ConversationChain( llm=llm, verbose=True, memory=ConversationBufferMemory() ) conversation.predict(input="How can one overcome anxiety?")

conversation.predict(input="Tell me more..")

Document Loader

LangChain allows seamlessly augmenting foundation models with custom datasets - like CSVs, Markdown, PDFs, etc. - to enable querying personalized and domain-specific information.

from langchain.document_loaders import TextLoader raw_document = TextLoader("/work/data/Gregory.txt").load()

Indexes

LangChain Indexes provide vector search over unstructured data - embedding text corpora into vectors, indexing them for efficient similarity lookup, enabling fast retrieval of relevant information to queries.

from langchain.embeddings.openai import OpenAIEmbeddings from langchain.text_splitter import CharacterTextSplitter from langchain.vectorstores import FAISS text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=0) documents = text_splitter.split_documents(raw_document) db = FAISS.from_documents(documents, OpenAIEmbeddings())

query = "When was Gregory born?" docs = db.similarity_search(query) print(docs[0].page_content)

Retrieval

LangChain enables combining database retrievers with a foundation model to return natural language responses to queries rather than just retrieving and displaying raw text from documents. This allows for generating more natural and conversational responses.

from langchain.chains import RetrievalQA from langchain.chat_models import ChatOpenAI llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0) qa_chain = RetrievalQA.from_chain_type(llm,retriever=db.as_retriever()) qa_chain({"query": "When was Gregory born?"})

Save & Load DBs

LangChain provides simple APIs for saving and loading different databases to enable persistence and reuse across sessions.

db.save_local("Gregory_index") new_db = FAISS.load_local("Gregory_index", OpenAIEmbeddings()) docs = new_db.similarity_search(query) print(docs[0].page_content)