Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,150 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import warnings
|
2 |
+
warnings.simplefilter(action='ignore', category=FutureWarning)
|
3 |
+
|
4 |
+
import PyPDF2
|
5 |
+
import gradio as gr
|
6 |
+
from langchain.prompts import PromptTemplate
|
7 |
+
from langchain.chains.summarize import load_summarize_chain
|
8 |
+
from huggingface_hub import login
|
9 |
+
from pathlib import Path
|
10 |
+
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint
|
11 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
12 |
+
import torch
|
13 |
+
import os
|
14 |
+
|
15 |
+
huggingface_token = os.getenv('HUGGINGFACE_TOKEN')
|
16 |
+
|
17 |
+
# Realizar el inicio de sesi贸n de Hugging Face solo si el token est谩 disponible
|
18 |
+
if huggingface_token:
|
19 |
+
login(token=huggingface_token)
|
20 |
+
|
21 |
+
# Configuraci贸n del modelo LLM
|
22 |
+
llm = HuggingFaceEndpoint(
|
23 |
+
repo_id="mistralai/Mistral-7B-Instruct-v0.3",
|
24 |
+
task="text-generation",
|
25 |
+
max_new_tokens=4096,
|
26 |
+
temperature=0.5,
|
27 |
+
do_sample=False,
|
28 |
+
)
|
29 |
+
llm_engine_hf = ChatHuggingFace(llm=llm)
|
30 |
+
|
31 |
+
# Configuraci贸n del modelo de clasificaci贸n
|
32 |
+
tokenizer = AutoTokenizer.from_pretrained("mrm8488/legal-longformer-base-8192-spanish")
|
33 |
+
model = AutoModelForSequenceClassification.from_pretrained("mrm8488/legal-longformer-base-8192-spanish")
|
34 |
+
|
35 |
+
id2label = {0: "multas", 1: "politicas_de_privacidad", 2: "contratos", 3: "denuncias", 4: "otros"}
|
36 |
+
|
37 |
+
def read_file(file):
|
38 |
+
file_path = file.name
|
39 |
+
if file_path.endswith('.pdf'):
|
40 |
+
return read_pdf(file_path)
|
41 |
+
else:
|
42 |
+
with open(file_path, 'r', encoding='utf-8') as f:
|
43 |
+
return f.read()
|
44 |
+
|
45 |
+
def read_pdf(file_path):
|
46 |
+
pdf_reader = PyPDF2.PdfReader(file_path)
|
47 |
+
text = ""
|
48 |
+
for page in range(len(pdf_reader.pages)):
|
49 |
+
text += pdf_reader.pages[page].extract_text()
|
50 |
+
return text
|
51 |
+
|
52 |
+
def summarize(text, summary_length):
|
53 |
+
if summary_length == 'Corto':
|
54 |
+
length_instruction = "El resumen debe tener un m谩ximo de 100 palabras."
|
55 |
+
elif summary_length == 'Medio':
|
56 |
+
length_instruction = "El resumen debe tener un m谩ximo de 500 palabras."
|
57 |
+
else:
|
58 |
+
length_instruction = "El resumen debe tener un m谩ximo de 1000 palabras."
|
59 |
+
|
60 |
+
template = f'''
|
61 |
+
Por favor, lea detenidamente el siguiente documento:
|
62 |
+
<document>
|
63 |
+
{{TEXT}}
|
64 |
+
</document>
|
65 |
+
Despu茅s de leer el documento, identifique los puntos clave y las ideas principales cubiertas en el texto. {length_instruction}
|
66 |
+
Su objetivo es ser exhaustivo en la captura del contenido central del documento, mientras que tambi茅n es conciso en la expresi贸n de cada punto del resumen. Omita los detalles menores y conc茅ntrese en los temas centrales y hechos importantes.
|
67 |
+
'''
|
68 |
+
|
69 |
+
prompt = PromptTemplate(
|
70 |
+
template=template,
|
71 |
+
input_variables=['TEXT']
|
72 |
+
)
|
73 |
+
|
74 |
+
formatted_prompt = prompt.format(TEXT=text)
|
75 |
+
output_summary = llm_engine_hf.invoke(formatted_prompt)
|
76 |
+
|
77 |
+
return output_summary.content
|
78 |
+
|
79 |
+
def classify_text(text):
|
80 |
+
inputs = tokenizer(text, return_tensors="pt", max_length=4096, truncation=True, padding="max_length")
|
81 |
+
model.eval()
|
82 |
+
with torch.no_grad():
|
83 |
+
outputs = model(**inputs)
|
84 |
+
logits = outputs.logits
|
85 |
+
predicted_class_id = logits.argmax(dim=-1).item()
|
86 |
+
predicted_label = id2label[predicted_class_id]
|
87 |
+
return f"Clasificaci贸n: {predicted_label}"
|
88 |
+
|
89 |
+
def translate(text, target_language):
|
90 |
+
template = '''
|
91 |
+
Por favor, traduzca el siguiente documento al {LANGUAGE}:
|
92 |
+
<document>
|
93 |
+
{TEXT}
|
94 |
+
</document>
|
95 |
+
Aseg煤rese de que la traducci贸n sea precisa y conserve el significado original del documento.
|
96 |
+
'''
|
97 |
+
|
98 |
+
prompt = PromptTemplate(
|
99 |
+
template=template,
|
100 |
+
input_variables=['TEXT', 'LANGUAGE']
|
101 |
+
)
|
102 |
+
|
103 |
+
formatted_prompt = prompt.format(TEXT=text, LANGUAGE=target_language)
|
104 |
+
translated_text = llm_engine_hf.invoke(formatted_prompt)
|
105 |
+
|
106 |
+
return translated_text.content
|
107 |
+
|
108 |
+
def process_file(file, action, target_language=None, summary_length=None):
|
109 |
+
text = read_file(file)
|
110 |
+
if action == "Resumen":
|
111 |
+
return summarize(text, summary_length)
|
112 |
+
elif action == "Clasificar":
|
113 |
+
return classify_text(text)
|
114 |
+
elif action == "Traducir":
|
115 |
+
return translate(text, target_language)
|
116 |
+
else:
|
117 |
+
return "Acci贸n no v谩lida"
|
118 |
+
|
119 |
+
# Crear la interfaz de Gradio
|
120 |
+
with gr.Blocks() as demo:
|
121 |
+
gr.Markdown("## LexAIcon Traducci贸n, Resumen y Clasificaci贸n")
|
122 |
+
gr.Image("icon.jpg", width=100)
|
123 |
+
|
124 |
+
with gr.Row():
|
125 |
+
with gr.Column():
|
126 |
+
file = gr.File(label="Subir un archivo")
|
127 |
+
action = gr.Radio(label="Seleccione una acci贸n", choices=["Resumen", "Clasificar", "Traducir"])
|
128 |
+
summary_length = gr.Radio(label="Seleccione la longitud del resumen", choices=["Corto", "Medio", "Largo"], visible=False)
|
129 |
+
target_language = gr.Dropdown(label="Seleccionar idioma de traducci贸n", choices=["en", "fr", "de"], visible=False)
|
130 |
+
|
131 |
+
with gr.Column():
|
132 |
+
output_text = gr.Textbox(label="Resultado", lines=60)
|
133 |
+
|
134 |
+
def update_ui(action):
|
135 |
+
if action == "Traducir":
|
136 |
+
return gr.update(visible=False), gr.update(visible=True)
|
137 |
+
elif action == "Resumen":
|
138 |
+
return gr.update(visible=True), gr.update(visible(False))
|
139 |
+
elif action == "Clasificar":
|
140 |
+
return gr.update(visible(False)), gr.update(visible=False)
|
141 |
+
else:
|
142 |
+
return gr.update(visible=False), gr.update(visible(False))
|
143 |
+
|
144 |
+
action.change(update_ui, inputs=action, outputs=[summary_length, target_language])
|
145 |
+
|
146 |
+
submit_button = gr.Button("Procesar")
|
147 |
+
submit_button.click(process_file, inputs=[file, action, target_language, summary_length], outputs=output_text)
|
148 |
+
|
149 |
+
# Ejecutar la aplicaci贸n Gradio
|
150 |
+
demo.launch(share=True)
|