app.py

import os
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_community.vectorstores import Chroma
from rerankers import Reranker

# Cargar PDF y partirlo en chunks
loader = PyPDFLoader("80dias.pdf")
documents = loader.load()
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=20)
splits = splitter.split_documents(documents)

# Crear embeddings
embedding_model = "sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
embeddings = HuggingFaceEmbeddings(model_name=embedding_model)
vectordb = Chroma.from_documents(splits, embedding=embeddings)

# Inicializar reranker
ranker = Reranker("answerdotai/answerai-colbert-small-v1", model_type="colbert")

# Cargar modelo de lenguaje de Hugging Face
model_id = "tiiuae/falcon-7b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype="auto")
generator = pipeline("text-generation", model=model, tokenizer=tokenizer)

# Función principal RAG
def rag_chat(message, history):
    query = message
    results = vectordb.similarity_search_with_score(query)
    
    # Seleccionar contextos relevantes
    context = []
    for doc, score in results:
        if score < 7:
            context.append(doc.page_content)

    if not context:
        return "No tengo información suficiente para responder a esa pregunta."

    # Aplicar reranking
    ranking = ranker.rank(query=query, docs=context)
    best_context = ranking[0].text

    # Crear prompt final
    prompt = f"""Responde a la siguiente pregunta utilizando solo el contexto proporcionado:

Contexto:
{best_context}

Pregunta: {query}
Respuesta:"""

    # Generar respuesta
    output = generator(prompt, max_new_tokens=100, do_sample=False)[0]["generated_text"]
    response = output.split("Respuesta:")[-1].strip()
    return response

# Gradio Chat Interface
iface = gr.ChatInterface(
    fn=rag_chat,
    title="Chat Julio Verne - RAG",
    description="Pregunta lo que quieras sobre *La vuelta al mundo en 80 días* de Julio Verne.",
    chatbot=gr.Chatbot(type="messages"),
    theme="default"
)

iface.launch()