app.py

import streamlit as st
import os
from PyPDF2 import PdfReader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings import HuggingFaceHubEmbeddings
from langchain.vectorstores import FAISS
from langchain.llms import HuggingFaceHub
from langchain.chains.question_answering import load_qa_chain
from langchain.prompts import PromptTemplate

st.set_page_config(page_title='preguntaDOC')
st.header("Pregunta a tu PDF")

# Campo para el token de Hugging Face (ahora requerido para los embeddings)
huggingface_api_token = st.text_input('Hugging Face API Token (requerido)', type='password')

pdf_obj = st.file_uploader("Carga tu documento", type="pdf", on_change=st.cache_resource.clear)

@st.cache_resource 
def create_embeddings(pdf, api_token):
    if not api_token:
        st.error("Se requiere un token de API de Hugging Face")
        return None
        
    os.environ["HUGGINGFACEHUB_API_TOKEN"] = api_token
    
    pdf_reader = PdfReader(pdf)
    text = ""
    for page in pdf_reader.pages:
        text += page.extract_text()
    
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=800,
        chunk_overlap=100,
        length_function=len
    )        
    chunks = text_splitter.split_text(text)
    
    # Usar HuggingFaceHubEmbeddings en lugar de HuggingFaceEmbeddings
    # Este enfoque no requiere sentence-transformers instalado localmente
    embeddings = HuggingFaceHubEmbeddings(
        repo_id="sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
        huggingfacehub_api_token=api_token
    )
    
    knowledge_base = FAISS.from_texts(chunks, embeddings)
    return knowledge_base

if pdf_obj and huggingface_api_token:
    knowledge_base = create_embeddings(pdf_obj, huggingface_api_token)
    
    if knowledge_base:
        user_question = st.text_input("Haz una pregunta sobre tu PDF:")
        
        if user_question:
            docs = knowledge_base.similarity_search(user_question, 3)
            
            # Usar un modelo gratuito de Hugging Face
            llm = HuggingFaceHub(
                repo_id="google/flan-t5-large",
                huggingfacehub_api_token=huggingface_api_token,
                model_kwargs={"temperature": 0.5, "max_length": 512}
            )
            
            prompt_template = """
            Responde a la siguiente pregunta basándote únicamente en el contexto proporcionado.
            
            Contexto: {context}
            
            Pregunta: {question}
            
            Respuesta:
            """
            
            PROMPT = PromptTemplate(
                template=prompt_template, 
                input_variables=["context", "question"]
            )
            
            chain = load_qa_chain(llm, chain_type="stuff", prompt=PROMPT)
            
            with st.spinner("Procesando tu pregunta..."):
                try:
                    respuesta = chain.run(input_documents=docs, question=user_question)
                    st.write(respuesta)
                except Exception as e:
                    st.error(f"Error al procesar tu pregunta: {str(e)}")
elif not huggingface_api_token and pdf_obj:
    st.warning("Por favor, ingresa tu token de API de Hugging Face para continuar.")