Python libray: langchain

framework for developing applications powered by language models

Open source framework that allows AI developers to combine Large Language Models (LLMs) like GPT-4 with external data.

Table of Contents

21 Jul 2023

Ecosystem

library-langchain/langchain-data-ecosystem.png

Integrations:

Document Loaders: 134
Vector Stores: 41
Embedding Models: 30
Chat Models: 8
LLMs: 54
Callbacks: 23
Tools: 87
Toolkits: 16

langchain.com
langchain.com
https://integrations.langchain.com/

LLM wrapper & unified API

Quickstart | 🦜️🔗 Langchain
langchain.com
https://python.langchain.com/docs/get_started/quickstart

Can't get basic example from documentation to work:

from langchain.llms import OpenAI

llm = OpenAI(openai_api_key=OPENAI_API_KEY, temperature=0.9)

llm.predict("What would be a good company name for a company that makes colorful socks?")

Error ⬇︎

llm.predict("What would be a good company name for a company that makes colorful socks?")
AttributeError: 'OpenAI' object has no attribute 'predict'

Same with other code example, copy pasted from documentation:

from langchain.prompts import PromptTemplate

prompt = PromptTemplate.from_template("What is a good name for a company that makes {product}?")
prompt.format(product="colorful socks")

Error ⬇︎

prompt = PromptTemplate.from_template("What is a good name for a company that makes {product}?")
AttributeError: type object 'PromptTemplate' has no attribute 'from_template'

🤔

ERROR SOLVED: pip install langchain --upgrade
Seemed I had installed langchain already a few months ago and was working off an old version 🙄

QA and Chat over Documents

Good overview:

Tutorial: ChatGPT Over Your Data
langchain.dev
https://blog.langchain.dev/tutorial-chatgpt-over-your-data/

Langchain OG:

QA and Chat over Documents | 🦜️🔗 Langchain
langchain.com
https://python.langchain.com/docs/use_cases/question_answering/

Code example for chat with code base:

Question Answering on Twitter algorithm codebase | 🦜️🔗 Langchain
langchain.com
https://python.langchain.com/docs/use_cases/code/twitter-the-algorithm-analysis-deeplake.html

Working code

My first working code - tested with long text files (1,000s of lines):

# Load document
from langchain.document_loaders import TextLoader
doc_loader = TextLoader('test.txt')
document = doc_loader.load()

# Split words
from langchain.text_splitter import RecursiveCharacterTextSplitter
text_splitter = RecursiveCharacterTextSplitter()
data = text_splitter.split_documents(document)

# Store embeddings
from langchain.vectorstores import Chroma
from langchain.embeddings import OpenAIEmbeddings
vectorstore = Chroma.from_documents(documents=data,embedding=OpenAIEmbeddings())

# Select Model
from langchain.chat_models import ChatOpenAI
llm = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0, openai_api_key=OPENAI_API_KEY)

# Create QA chain
from langchain.chains import RetrievalQA
qa_chain = RetrievalQA.from_chain_type(llm,retriever=vectorstore.as_retriever())

# Ask question
question = """
        Tell me what is in this text.
"""

# Output
output = qa_chain({"query": question})

print(f"\n{output['result']}\n")

links

social