RAG, retrieval augmented generation, con documentos .pdf, LLM phi3 local y Ollama.

from langchain_community.document_loaders import UnstructuredPDFLoader
from langchain_community.embeddings import OllamaEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma
from langchain.prompts import ChatPromptTemplate, PromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_community.chat_models import ChatOllama
from langchain_core.runnables import RunnablePassthrough
from langchain.retrievers.multi_query import MultiQueryRetriever

# Librerias necesarias:
#pip3 install unstructured langchain
#pip3 install "unstructured[all-docs]"
#pip3 install chromadb
#pip3 install langchain-text-splitters
#pip3 install langchain_community

local_path = "esp32manual.pdf"

# Local PDF file uploads
if local_path:
  loader = UnstructuredPDFLoader(file_path=local_path)
  data = loader.load()
else:
  print("Upload a PDF file")
  
# Preview first page
data[0].page_content

# Split and chunk 
text_splitter = RecursiveCharacterTextSplitter(chunk_size=7500, chunk_overlap=100)
chunks = text_splitter.split_documents(data)

# Add to vector database
vector_db = Chroma.from_documents(
    documents=chunks, 
    embedding=OllamaEmbeddings(model="phi3",show_progress=True),
    collection_name="local-rag"
)

# LLM from Ollama
local_model = "phi3"
llm = ChatOllama(model=local_model)

QUERY_PROMPT = PromptTemplate(
    input_variables=["question"],
    template="""Eres un asistente de modelo de lenguaje de IA. 
    Su tarea principal es responder a mis preguntas.
    Pregunta original: {question}""",
)

retriever = MultiQueryRetriever.from_llm(
    vector_db.as_retriever(), 
    llm,
    prompt=QUERY_PROMPT
)

# RAG prompt
template = """Las respuestas estan basadas para asistir al usuario en diferentes contxtos:
{context}
Pregunta: {question}
"""

prompt = ChatPromptTemplate.from_template(template)

chain = (
    {"context": retriever, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

# Bucle para continuar preguntando hasta que el usuario escriba "adiós"
while True:
    pregunta_usuario = input("Cesar: ")
    if pregunta_usuario.lower() == "adios":
        break
    resultado = chain.invoke(pregunta_usuario)
    print(resultado)

# Delete all collections in the db
vector_db.delete_collection()
print("Hasta pronto")