langchain_backend/utils.py

from langchain_community.document_loaders import PyPDFLoader
import os
from langchain_openai import ChatOpenAI
from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain_core.prompts import ChatPromptTemplate
from langchain_huggingface import HuggingFaceEndpoint
from setup.environment import default_model

os.environ.get("OPENAI_API_KEY")
os.environ.get("HUGGINGFACEHUB_API_TOKEN")

def getPDF(file_path="./nike.pdf"):
  text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
  loader = PyPDFLoader(file_path, extract_images=False)
  pages = loader.load_and_split(text_splitter)
  return pages

def create_retriever(documents):
  vectorstore = Chroma.from_documents(
    documents,
    embedding=OpenAIEmbeddings(),
)

  retriever = vectorstore.as_retriever(
      search_type="similarity",
      search_kwargs={"k": 1},
  )
  
  return retriever

def create_prompt_llm_chain(system_prompt, modelParam):
  if modelParam == default_model:
    model = ChatOpenAI(model=modelParam)
  else:
    model = HuggingFaceEndpoint(
        repo_id=modelParam,
        task="text-generation",
        max_new_tokens=100,
        do_sample=False,
        huggingfacehub_api_token=os.environ.get("HUGGINGFACEHUB_API_TOKEN")
    )
    # result = model.invoke("Hugging Face is")
    # print('result: ', result)

  system_prompt = system_prompt + "\n\n" + "{context}"
  prompt = ChatPromptTemplate.from_messages(
      [
          ("system", system_prompt),
          ("human", "{input}"),
      ]
  )
  question_answer_chain = create_stuff_documents_chain(model, prompt)
  return question_answer_chain