Spaces:
Running
Running
import streamlit as st | |
import gradio as gr | |
from PyPDF2 import PdfReader | |
from langchain.text_splitter import RecursiveCharacterTextSplitter | |
import os | |
from langchain_community.embeddings import HuggingFaceEmbeddings | |
from langchain.vectorstores import FAISS | |
from langchain_groq import ChatGroq | |
from langchain.chains.question_answering import load_qa_chain | |
from langchain.prompts import PromptTemplate | |
from dotenv import load_dotenv | |
import re | |
# Load environment variables | |
load_dotenv() | |
os.getenv("GROQ_API_KEY") | |
def get_pdf_text(pdf_docs): | |
"""Extrae texto de los archivos PDF cargados.""" | |
text = "" | |
for pdf in pdf_docs: | |
pdf_reader = PdfReader(pdf) | |
for page in pdf_reader.pages: | |
text += page.extract_text() | |
return text | |
def get_text_chunks(text): | |
"""Divide el texto extra铆do en fragmentos manejables.""" | |
text_splitter = RecursiveCharacterTextSplitter(chunk_size=5000, chunk_overlap=500) | |
chunks = text_splitter.split_text(text) | |
return chunks | |
def get_vector_store(text_chunks): | |
"""Crea y guarda un almac茅n de vectores FAISS a partir de fragmentos de texto.""" | |
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2") | |
vector_store = FAISS.from_texts(text_chunks, embedding=embeddings) | |
vector_store.save_local("faiss_index") | |
def get_conversational_chain(): | |
"""Configura una cadena conversacional usando el modelo Groq LLM.""" | |
prompt_template = """ | |
Responde la pregunta en espa帽ol de la manera m谩s detallada posible a partir del contexto proporcionado. Si la respuesta no est谩 en | |
el contexto proporcionado, simplemente di, "la respuesta no est谩 disponible en el contexto." No proporciones respuestas incorrectas. | |
Contexto: | |
{context}? | |
Pregunta: | |
{question} | |
Respuesta: | |
""" | |
model = ChatGroq( | |
temperature=0.3, | |
model_name="deepseek-r1-distill-llama-70b", | |
groq_api_key=os.getenv("GROQ_API_KEY") | |
) | |
prompt = PromptTemplate(template=prompt_template, input_variables=["context", "question"]) | |
chain = load_qa_chain(model, chain_type="stuff", prompt=prompt) | |
return chain | |
def eliminar_texto_entre_tags(texto): | |
patron = r'<think>.*?</think>' | |
texto_limpio = re.sub(patron, '', texto, flags=re.DOTALL) | |
return texto_limpio | |
def user_input(user_question): | |
"""Maneja las consultas del usuario recuperando respuestas del almac茅n de vectores.""" | |
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2") | |
new_db = FAISS.load_local("faiss_index", embeddings, allow_dangerous_deserialization=True) | |
docs = new_db.similarity_search(user_question) | |
chain = get_conversational_chain() | |
response = chain( | |
{"input_documents": docs, "question": user_question}, | |
return_only_outputs=True | |
) | |
# Depuraci贸n: Imprimir la respuesta original | |
original_response = response['output_text'] | |
print("Original Response:", original_response) | |
# Extraer el proceso de pensamiento | |
thought_process = "" | |
if "<think>" in response['output_text'] and "</think>" in response['output_text']: | |
thought_process_match = re.search(r"<think>(.*?)</think>", response['output_text'], re.DOTALL) | |
if thought_process_match: | |
thought_process = thought_process_match.group(1).strip() | |
# Eliminar el proceso de pensamiento de la respuesta principal | |
clean_response = eliminar_texto_entre_tags(original_response) | |
# Depuraci贸n: Imprimir la respuesta limpia | |
print("Cleaned Response:", clean_response) | |
# Mostrar el proceso de pensamiento del modelo en el expander | |
with st.expander("Proceso de Pensamiento del Modelo"): | |
st.write(thought_process) | |
st.markdown(f"### Respuesta:\n{clean_response}") | |
def main(): | |
"""Funci贸n principal para ejecutar la aplicaci贸n Streamlit.""" | |
st.set_page_config(page_title="Chat PDF", page_icon=":books:", layout="wide") | |
# Configuraci贸n de la apariencia de la aplicaci贸n | |
st.markdown( | |
""" | |
<style> | |
body { | |
background-color: #1E90FF; | |
color: white; | |
} | |
.sidebar .sidebar-content { | |
background-color: #00008B; | |
} | |
.main { | |
background-color: #00008B; | |
color: white; | |
} | |
.stButton>button { | |
background-color: #0b0175; | |
color: white; | |
} | |
/* Estilos personalizados para los botones espec铆ficos */ | |
.custom-button button { | |
background-color: transparent; | |
border: 2px solid gray; | |
color: gray; | |
transition: all 0.3s ease; | |
margin-right: 10px; /* Espacio entre botones */ | |
} | |
.custom-button button:hover { | |
background-color: gray; | |
color: white; | |
} | |
.custom-button { | |
display: flex; | |
gap: 10px; /* Espacio entre botones */ | |
} | |
</style> | |
""", | |
unsafe_allow_html=True | |
) | |
st.title("PDF Consultor") | |
with st.sidebar: | |
pdf_docs = st.file_uploader( | |
"Subir archivo PDF", | |
accept_multiple_files=True, | |
type=["pdf"] | |
) | |
if st.button("Procesar"): | |
with st.spinner("Procesando el archivo..."): | |
raw_text = get_pdf_text(pdf_docs) | |
text_chunks = get_text_chunks(raw_text) | |
get_vector_store(text_chunks) | |
st.success("隆PDF procesado exitosamente!") | |
# Botones para preguntas predefinidas con estilo personalizado | |
col1, col2, col3 = st.columns(3) | |
with col1: | |
st.markdown('<div class="custom-button">', unsafe_allow_html=True) | |
button1 = st.button("Resumen", key="resumen_button") | |
st.markdown('</div>', unsafe_allow_html=True) | |
with col2: | |
st.markdown('<div class="custom-button">', unsafe_allow_html=True) | |
button2 = st.button("Entidad", key="entidad_button") | |
st.markdown('</div>', unsafe_allow_html=True) | |
with col3: | |
st.markdown('<div class="custom-button">', unsafe_allow_html=True) | |
button3 = st.button("Fecha implantaci贸n", key="fecha_button") | |
st.markdown('</div>', unsafe_allow_html=True) | |
if button1: | |
# Do something... | |
user_input("Realiza un resumen sobre los aspectos m谩s relevantes comentados en el documento") | |
if button2: | |
# Do something... | |
user_input("A qu茅 entidad pertenece el contenido del documento?") | |
if button3: | |
# Do something... | |
user_input("En qu茅 fecha se implantar谩 el contenido del documento?") | |
user_question = st.text_input("Introduce tu pregunta", placeholder="驴Qu茅 quieres saber?") | |
if user_question: | |
with st.spinner("Obteniendo tu respuesta..."): | |
user_input(user_question) | |
if __name__ == "__main__": | |
main() | |