import gradio as gr
import requests

from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.document_loaders import WebBaseLoader
from langchain_community.vectorstores import Chroma
from langchain_huggingface import HuggingFaceEmbeddings
from langchain.document_loaders import PyPDFLoader
from rerankers import Reranker
from langchain_community.chat_models import ChatOllama
from langchain import hub
from langchain_core.output_parsers import StrOutputParser

# LLM local
local_llm = "llama3.2:1b"
llm = ChatOllama(model=local_llm, temperature=0, top_k=50, top_p=0.95)
chain = llm | StrOutputParser()

# Descargar y cargar PDF
URL = "https://rua.ua.es/dspace/bitstream/10045/118060/1/Isurus_01_02.pdf"
response = requests.get(URL)
open("Isurus_01_02.pdf", "wb").write(response.content)

loader = PyPDFLoader("Isurus_01_02.pdf")
documents = loader.load()

# División en chunks
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=20)
all_splits = text_splitter.split_documents(documents)

# Embeddings con device CPU
model_name = "sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
model_kwargs = {"device": "cpu"}
embeddings = HuggingFaceEmbeddings(model_name=model_name, model_kwargs=model_kwargs)

# Vector store
vectordb = Chroma.from_documents(documents=all_splits, embedding=embeddings, persist_directory="chroma_db")

# Función para test RAG
def test_rag(query):
    docs = vectordb.similarity_search_with_score(query)
    prompt = hub.pull("rlm/rag-prompt")
    rag_chain = prompt | llm | StrOutputParser()

    context = []
    for doc, score in docs:
        if score < 7:
            doc_details = doc.to_json()['kwargs']
            context.append(doc_details['page_content'])

    if context:
        generation = rag_chain.invoke({"context": "\n".join(context), "question": query})
        return generation
    else:
        return "No tengo información para responder a esta pregunta."

# Interfaz de Gradio
def chat_function(message, history):
    return test_rag(message)

iface = gr.ChatInterface(
    fn=chat_function,
    chatbot=gr.Chatbot(),
    title="RAG Bot",
    description="Hazme preguntas sobre el contenido del PDF.",
)

iface.launch()