Spaces:
Sleeping
Sleeping
File size: 2,310 Bytes
514c042 c77261f 514c042 6dbb17f 514c042 9b39e4a 514c042 179639e 514c042 179639e 514c042 179639e 514c042 179639e 514c042 179639e 514c042 179639e 514c042 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 |
import gradio as gr
from langchain_community.llms import HuggingFaceHub
from langchain_community.vectorstores import Chroma
from langchain_community.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_core.output_parsers import StrOutputParser
from langchain_huggingface import HuggingFaceEmbeddings
from langchain import hub
from rerankers import Reranker
import os
# Configuraci贸n del token de acceso a Hugging Face (si usas modelo privado)
os.environ["HUGGINGFACEHUB_API_TOKEN"] = os.getenv("HUGGINGFACEHUB_API_TOKEN")
# Cargar PDF
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)
# Modelo LLM desde HuggingFace (usa uno disponible en Spaces)
llm = HuggingFaceHub(repo_id="mistralai/Mistral-7B-Instruct-v0.1", model_kwargs={"temperature": 0.5, "max_new_tokens": 500})
chain = llm | StrOutputParser()
# Reranker
ranker = Reranker("answerdotai/answerai-colbert-small-v1", model_type="colbert")
# Funci贸n RAG
def rag_chat(message, history):
# Solo usamos el mensaje del usuario
query = message
results = vectordb.similarity_search_with_score(query)
context = []
for doc, score in results:
if score < 7:
context.append(doc.page_content)
if not context:
return "No tengo informaci贸n suficiente para responder a esa pregunta."
ranking = ranker.rank(query=query, docs=context)
best_context = ranking[0].text
prompt = f"""Contesta a la siguiente pregunta usando solo el contexto que se proporciona:
Contexto:
{best_context}
Pregunta: {query}
Respuesta:"""
return llm.invoke(prompt)
# Interfaz Gradio
iface = gr.ChatInterface(
fn=rag_chat,
title="Chat Julio Verne - RAG",
description="Pregunta lo que quieras sobre *La vuelta al mundo en 80 d铆as* de Julio Verne.",
chatbot=gr.Chatbot(type="messages") # 馃憟 Esto elimina el warning de formato obsoleto
)
iface.launch()
|