Spaces:
Runtime error
Runtime error
Jose Alvaro Luna G
commited on
Commit
路
c5446a2
1
Parent(s):
e5da8b5
feat: app init
Browse files- extract_text.py +21 -0
- main.py +99 -4
- requirements.txt +9 -0
extract_text.py
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# extract_text.py
|
| 2 |
+
from pdfminer.high_level import extract_text
|
| 3 |
+
from docx import Document
|
| 4 |
+
import pytesseract
|
| 5 |
+
from PIL import Image
|
| 6 |
+
|
| 7 |
+
def extract_text_from_image(file_path):
|
| 8 |
+
image = Image.open(file_path)
|
| 9 |
+
text = pytesseract.image_to_string(image)
|
| 10 |
+
return text
|
| 11 |
+
|
| 12 |
+
def extract_text_from_docx(file_path):
|
| 13 |
+
doc = Document(file_path)
|
| 14 |
+
full_text = []
|
| 15 |
+
for para in doc.paragraphs:
|
| 16 |
+
full_text.append(para.text)
|
| 17 |
+
return '\n'.join(full_text)
|
| 18 |
+
|
| 19 |
+
def extract_text_from_pdf(file_path):
|
| 20 |
+
text = extract_text(file_path)
|
| 21 |
+
return text
|
main.py
CHANGED
|
@@ -1,7 +1,102 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
|
| 3 |
-
|
| 4 |
-
|
| 5 |
|
| 6 |
-
|
| 7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# app.py
|
| 2 |
+
import torch
|
| 3 |
+
from transformers import (
|
| 4 |
+
DPRContextEncoder, DPRContextEncoderTokenizerFast,
|
| 5 |
+
DPRQuestionEncoder, DPRQuestionEncoderTokenizerFast,
|
| 6 |
+
BartForConditionalGeneration, BartTokenizer
|
| 7 |
+
)
|
| 8 |
+
from datasets import Dataset
|
| 9 |
+
import faiss
|
| 10 |
+
import numpy as np
|
| 11 |
import gradio as gr
|
| 12 |
|
| 13 |
+
# Importar funciones de extracci贸n
|
| 14 |
+
from extract_text import extract_text_from_pdf, extract_text_from_docx, extract_text_from_image
|
| 15 |
|
| 16 |
+
# Inicializar modelos y variables globales
|
| 17 |
+
ctx_encoder = DPRContextEncoder.from_pretrained('facebook/dpr-ctx_encoder-single-nq-base')
|
| 18 |
+
ctx_tokenizer = DPRContextEncoderTokenizerFast.from_pretrained('facebook/dpr-ctx_encoder-single-nq-base')
|
| 19 |
+
|
| 20 |
+
q_encoder = DPRQuestionEncoder.from_pretrained('facebook/dpr-question_encoder-single-nq-base')
|
| 21 |
+
q_tokenizer = DPRQuestionEncoderTokenizerFast.from_pretrained('facebook/dpr-question_encoder-single-nq-base')
|
| 22 |
+
|
| 23 |
+
generator = BartForConditionalGeneration.from_pretrained('facebook/bart-large')
|
| 24 |
+
gen_tokenizer = BartTokenizer.from_pretrained('facebook/bart-large')
|
| 25 |
+
|
| 26 |
+
# Inicializar dataset y 铆ndice
|
| 27 |
+
dataset = Dataset.from_dict({'text': []})
|
| 28 |
+
embeddings = np.empty((0, ctx_encoder.config.hidden_size), dtype='float32')
|
| 29 |
+
index = faiss.IndexFlatIP(ctx_encoder.config.hidden_size)
|
| 30 |
+
|
| 31 |
+
# Funci贸n para actualizar el 铆ndice con nuevo texto
|
| 32 |
+
def actualizar_indice(nuevo_texto):
|
| 33 |
+
global dataset, embeddings, index
|
| 34 |
+
|
| 35 |
+
# A帽adir nuevo documento al dataset
|
| 36 |
+
dataset = dataset.add_item({'text': nuevo_texto})
|
| 37 |
+
|
| 38 |
+
# Codificar el nuevo documento
|
| 39 |
+
inputs = ctx_tokenizer(nuevo_texto, truncation=True, padding='longest', return_tensors='pt')
|
| 40 |
+
embedding = ctx_encoder(**inputs).pooler_output.detach().numpy()
|
| 41 |
+
|
| 42 |
+
# Actualizar embeddings y 铆ndice
|
| 43 |
+
embeddings = np.vstack([embeddings, embedding])
|
| 44 |
+
index.add(embedding)
|
| 45 |
+
|
| 46 |
+
# Funci贸n para recuperar documentos relevantes
|
| 47 |
+
def retrieve_docs(question, k=5):
|
| 48 |
+
inputs = q_tokenizer(question, return_tensors='pt')
|
| 49 |
+
question_embedding = q_encoder(**inputs).pooler_output.detach().numpy()
|
| 50 |
+
|
| 51 |
+
distances, indices = index.search(question_embedding, k)
|
| 52 |
+
retrieved_texts = [dataset[i]['text'] for i in indices[0]]
|
| 53 |
+
return retrieved_texts
|
| 54 |
+
|
| 55 |
+
# Funci贸n para generar respuesta
|
| 56 |
+
def generate_answer(question):
|
| 57 |
+
retrieved_docs = retrieve_docs(question)
|
| 58 |
+
context = ' '.join(retrieved_docs)
|
| 59 |
+
|
| 60 |
+
input_text = f"Pregunta: {question} Contexto: {context}"
|
| 61 |
+
inputs = gen_tokenizer([input_text], max_length=1024, return_tensors='pt', truncation=True)
|
| 62 |
+
summary_ids = generator.generate(inputs['input_ids'], num_beams=4, max_length=100, early_stopping=True)
|
| 63 |
+
answer = gen_tokenizer.decode(summary_ids[0], skip_special_tokens=True)
|
| 64 |
+
return answer
|
| 65 |
+
|
| 66 |
+
# Funci贸n principal de la aplicaci贸n
|
| 67 |
+
def responder(archivo, pregunta):
|
| 68 |
+
texto_extraido = ''
|
| 69 |
+
if archivo is not None:
|
| 70 |
+
file_path = archivo.name
|
| 71 |
+
if file_path.endswith('.pdf'):
|
| 72 |
+
texto_extraido = extract_text_from_pdf(file_path)
|
| 73 |
+
elif file_path.endswith('.docx'):
|
| 74 |
+
texto_extraido = extract_text_from_docx(file_path)
|
| 75 |
+
elif file_path.lower().endswith(('.png', '.jpg', '.jpeg')):
|
| 76 |
+
texto_extraido = extract_text_from_image(file_path)
|
| 77 |
+
else:
|
| 78 |
+
return "Formato de archivo no soportado."
|
| 79 |
+
|
| 80 |
+
# Actualizar el 铆ndice con el nuevo texto
|
| 81 |
+
actualizar_indice(texto_extraido)
|
| 82 |
+
|
| 83 |
+
# Generar respuesta
|
| 84 |
+
respuesta = generate_answer(pregunta)
|
| 85 |
+
return respuesta
|
| 86 |
+
else:
|
| 87 |
+
return "Por favor, sube un archivo."
|
| 88 |
+
|
| 89 |
+
# Configurar la interfaz de Gradio
|
| 90 |
+
interfaz = gr.Interface(
|
| 91 |
+
fn=responder,
|
| 92 |
+
inputs=[
|
| 93 |
+
gr.inputs.File(label="Sube un archivo (PDF, DOCX, Imagen)"),
|
| 94 |
+
gr.inputs.Textbox(lines=2, placeholder="Escribe tu pregunta aqu铆...")
|
| 95 |
+
],
|
| 96 |
+
outputs="text",
|
| 97 |
+
title="Aplicaci贸n RAG con Extracci贸n de Texto",
|
| 98 |
+
description="Sube un archivo y haz una pregunta sobre su contenido."
|
| 99 |
+
)
|
| 100 |
+
|
| 101 |
+
if __name__ == "__main__":
|
| 102 |
+
interfaz.launch()
|
requirements.txt
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
transformers
|
| 2 |
+
datasets
|
| 3 |
+
faiss-cpu
|
| 4 |
+
gradio
|
| 5 |
+
pytesseract
|
| 6 |
+
Pillow
|
| 7 |
+
pdfminer.six
|
| 8 |
+
python-docx
|
| 9 |
+
torch
|