AI - LangChain

Introduction

LangChain is an open source framework launched in October 2022 that allows AI developers to combine AI Large Language Models (LLMs) with external sources of computation and data (eg. your proprietary private documents or database) and then perform an output task as a response or as an email
In April 2023, LangChain had incorporated and had quickly raised $US30m in seed funding
available in Python, JavaScript, TypeScript
your documents are embedded into a vector database
examples of usage:
- AI - creating a voiced role-play conversational output from ChatGTP

There are 3 main concepts

components:
- LLM wrappers
- prompt templates
- indexes for relevant information retrieval
chains:
- these assemble components to solve a specific task
agents:
- allow LLMs to interact with its environment eg. via APIs

Python coding example

from https://www.youtube.com/watch?v=aywZrzNaKjs
assumes you have set up accounts with OpenAI and PineCone and have the API_KEYS etc for each
assumes you have set your API keys etc in dotenv file
assumes you have install the libraries dotenv, langchain, pinecone

from dotenv import load_dotenv, find_dotenv
load dot_env(find_dotenv)

from lanchain.llms import OpenAI
llm = OpenAI(model_name="text-davinci-003")
llm("explain large language models in one sentence")

from langchain.schema import (
    AIMessage,
    HumanMessage,
    SystemMessage
)

from langchain.chat_models import ChatOpenAI

chat = ChatOpenAI(model_name="gpt-3.5-turbp",temperature=0.3)
messages = [
     SystemMessage(content="You are an expert data scientist")
     HumanMessage(content="Write a Python script that trains a neural network on data")
]
response = chat(messages)
# now see the output to ensure it works so far:
print(response.content,end='\n')

# now create mechanism for dynamic prompts:

from langchain import PromptTemplate

template = """
You are an expert data scientist with an expertise in building deep learning models.
Explain the concept of {concept} in a couple of lines
"""

prompt = PromptTemplate(
   input_variables=["concept"],
   template=template,
)

llm(prompt.format(concept="regularization"))

#now add in the chain:

from langchain.chains import LLMChain
chain = LLMChain(llm=llm, prompt=prompt)

#run the chain only specifying the input variable:
print(chain.run("regularization"))

#now lets chain that output to a LLM input:
second_prompt = PromptTemplate(
   input_variables=["ml_concept"],
   template="Turn the concept description of {ml_concept} and explain it to me like I'm five in 500 words",
)

chain_two = LLMChain(llm=llm, prompt=second_prompt)

#combine the chains:
from langchain.chains import SimpleSequentialChain
overall_chain = SimpleSequentialChain(chains=[chain, chain_two], verbose=True)

#now run it:
explanation = overall_chain.run("regularization")
print(explanation)

#now take this response and split it into chunks for storage in a Pinecone vector embedding datastore:
from langchain.text_splitter import RecursiveCharacterTextSplitter

text_splitter = RecursiveCharacterTextSplitter(
     chunk_size = 100,
     chunk_overlap = 0,
)

texts = text_splitter.create_documents([explanation])

#display 1st line via:
texts[0].page_content

#now embed text:
from langchain.embeddings import OpenAIEmbeddings

embeddings = OpenAIEmbeddings(model_name="ada")

#can view vector embeddings output:
query_result = embeddings.embed_query(texts[0].page_content)
query_result

#now store in PineCone:
import os
import pinecone
from langchain.vector_stores import Pinecone

#initialize Pinecone:
pinecone.init(
     api_key=os.getenv("PINECONE_API_KEY"),
     environment=os.getenv("PINECONE_ENV")
)

#store embeddings
index_name = "langchain-quickstart"
search = Pinecone.from_documents(texts, embeddings, index_name=index_name)

#query pinecone:
query = "What is magical about a regularizer?"
result = search.similarity_search(query)

#now create a langchain agent to have a LLM run Python code to solve natural language questions:
from langchain.agents.agent_toolkits import create_python_agent
from langchain.tools.python.tool import PythonREPLTool
from langchain.python import PythonREPL
from langchain.llms.openai import OpenAI

#create agent to get LLM to run Python code from natural language:
agent_executor = create_python_agent(
     llm=OpenAI(temperature=0, max_tokens=1000),
     tool=PythonREPLTool(),
     verbose=True
)

#example:
agent_executor.run("Find the roots (zeros) of the quadratic function 3 * x**2 + 2*x -1")

OzEMedicine - Wiki for Australian Emergency Medicine Doctors

Table of Contents

AI - LangChain

Introduction

There are 3 main concepts

Python coding example