_utils/main.py

import os
from _utils.utils import create_prompt_llm_chain, create_retriever, getPDF, create_llm, create_prompt_llm_chain_summary, process_embedding_summary
from _utils import utils
from langchain.chains import create_retrieval_chain
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings
from langchain.chains.summarize import load_summarize_chain

os.environ.get("OPENAI_API_KEY")

def get_llm_answer(system_prompt, user_prompt, pdf_url, model, embedding):
  if embedding == "gpt":
    embedding_object = OpenAIEmbeddings()
  else:
    embedding_object = HuggingFaceEmbeddings(model_name=embedding)

  vectorstore = Chroma(
      collection_name="documents",
      embedding_function=embedding_object
    )

  print('model: ', model)
  print('embedding: ', embedding)
  pages = []
  if pdf_url:
    pages = getPDF(pdf_url)
  else:
    pages = getPDF()
  retriever = create_retriever(pages, vectorstore)
  rag_chain = create_retrieval_chain(retriever, create_prompt_llm_chain(system_prompt, model))
  results = rag_chain.invoke({"input": user_prompt})
  # print('allIds ARQUIVO MAIN: ',  utils.allIds)
  vectorstore.delete( utils.allIds)
  vectorstore.delete_collection()
  utils.allIds = []
  # print('utils.allIds: ', utils.allIds)
  return results

def get_llm_answer_summary(system_prompt, user_prompt, pdf_url, model, isIterativeRefinement):
  print('model: ', model)
  print('isIterativeRefinement: ', isIterativeRefinement)
  print('\n\n\n')
  pages = getPDF(pdf_url)
  if not isIterativeRefinement:
    rag_chain = create_prompt_llm_chain_summary(system_prompt, model)

    results = rag_chain.invoke({"input": user_prompt, "context": pages})

    return results
  else:
    chain = load_summarize_chain(create_llm(model), "refine", True)
    result = chain.invoke({"input_documents": pages})
    print('result: ', result)
    return result
    # Obs --> Para passar informações personalizadas --> chain = load_summarize_chain(llm, "refine", True, question_prompt=initial_prompt, refine_prompt=PromptTemplate.from_template(refine_prompt))
    # Para ver mais opções --> Acessa a origem da função load_summarize_chain , e nela acessa a origem da função _load_refine_chain --> As opções são os parâmetros que esta última função recebe

def get_llm_answer_summary_with_embedding(system_prompt, user_prompt, pdf_url, model, isIterativeRefinement):
  print('model: ', model)
  print('isIterativeRefinement: ', isIterativeRefinement)
  print('\n\n\n')
  pages = getPDF(pdf_url)
  full_texto = ""
  for p in pages:
    full_texto += p.page_content
  print('full_texto: ', full_texto)

  rag_chain = process_embedding_summary(system_prompt, model)

  results = rag_chain.invoke({"input": user_prompt, "context": pages})

  return results