Update app.py
Browse files
app.py
CHANGED
|
@@ -14,7 +14,7 @@ import os
|
|
| 14 |
|
| 15 |
login(token=os.getenv('HUGGINGFACE_TOKEN'))
|
| 16 |
|
| 17 |
-
# Configuraci贸n del modelo
|
| 18 |
llm = HuggingFaceEndpoint(
|
| 19 |
repo_id="mistralai/Mistral-7B-Instruct-v0.3",
|
| 20 |
task="text-generation",
|
|
@@ -37,7 +37,7 @@ def read_pdf(file_path):
|
|
| 37 |
text += pdf_reader.pages[page].extract_text()
|
| 38 |
return text
|
| 39 |
|
| 40 |
-
def summarize(file):
|
| 41 |
# Leer el contenido del archivo subido
|
| 42 |
file_path = file.name
|
| 43 |
if file_path.endswith('.pdf'):
|
|
@@ -45,14 +45,21 @@ def summarize(file):
|
|
| 45 |
else:
|
| 46 |
with open(file_path, 'r', encoding='utf-8') as f:
|
| 47 |
text = f.read()
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
<document>
|
| 52 |
-
{TEXT}
|
| 53 |
</document>
|
| 54 |
-
|
| 55 |
-
|
| 56 |
'''
|
| 57 |
|
| 58 |
prompt = PromptTemplate(
|
|
@@ -63,7 +70,7 @@ Your goal is to be comprehensive in capturing the core content of the document,
|
|
| 63 |
formatted_prompt = prompt.format(TEXT=text)
|
| 64 |
output_summary = llm_engine_hf.invoke(formatted_prompt)
|
| 65 |
|
| 66 |
-
return output_summary.content
|
| 67 |
|
| 68 |
def classify_text(text):
|
| 69 |
inputs = tokenizer(text, return_tensors="pt", max_length=4096, truncation=True, padding="max_length")
|
|
@@ -85,11 +92,11 @@ def translate(file, target_language):
|
|
| 85 |
text = f.read()
|
| 86 |
|
| 87 |
template = '''
|
| 88 |
-
|
| 89 |
<document>
|
| 90 |
{TEXT}
|
| 91 |
</document>
|
| 92 |
-
|
| 93 |
'''
|
| 94 |
|
| 95 |
prompt = PromptTemplate(
|
|
@@ -100,11 +107,11 @@ Ensure that the translation is accurate and preserves the original meaning of th
|
|
| 100 |
formatted_prompt = prompt.format(TEXT=text, LANGUAGE=target_language)
|
| 101 |
translated_text = llm_engine_hf.invoke(formatted_prompt)
|
| 102 |
|
| 103 |
-
return translated_text.content
|
| 104 |
|
| 105 |
-
def process_file(file, action, target_language=None):
|
| 106 |
if action == "Resumen":
|
| 107 |
-
return summarize(file)
|
| 108 |
elif action == "Clasificar":
|
| 109 |
file_path = file.name
|
| 110 |
if file_path.endswith('.pdf'):
|
|
@@ -118,6 +125,15 @@ def process_file(file, action, target_language=None):
|
|
| 118 |
else:
|
| 119 |
return "Acci贸n no v谩lida"
|
| 120 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 121 |
def download_text(output_text, filename='output.txt'):
|
| 122 |
if output_text:
|
| 123 |
file_path = Path(filename)
|
|
@@ -133,27 +149,34 @@ def create_download_file(output_text, filename='output.txt'):
|
|
| 133 |
|
| 134 |
# Crear la interfaz de Gradio
|
| 135 |
with gr.Blocks() as demo:
|
| 136 |
-
gr.Markdown("##
|
| 137 |
|
| 138 |
with gr.Row():
|
| 139 |
with gr.Column():
|
| 140 |
file = gr.File(label="Subir un archivo")
|
| 141 |
action = gr.Radio(label="Seleccione una acci贸n", choices=["Resumen", "Clasificar", "Traducir"])
|
| 142 |
target_language = gr.Dropdown(label="Seleccionar idioma de traducci贸n", choices=["en", "fr", "de"], visible=False)
|
|
|
|
| 143 |
|
| 144 |
with gr.Column():
|
| 145 |
output_text = gr.Textbox(label="Resultado", lines=20)
|
|
|
|
|
|
|
| 146 |
|
| 147 |
-
def
|
| 148 |
if action == "Traducir":
|
| 149 |
-
return gr.update(visible=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 150 |
else:
|
| 151 |
-
return gr.update(visible=False)
|
| 152 |
|
| 153 |
-
action.change(
|
| 154 |
|
| 155 |
submit_button = gr.Button("Procesar")
|
| 156 |
-
submit_button.click(process_file, inputs=[file, action, target_language], outputs=output_text)
|
| 157 |
|
| 158 |
def generate_file():
|
| 159 |
summary_text = output_text.value
|
|
@@ -168,5 +191,8 @@ with gr.Blocks() as demo:
|
|
| 168 |
outputs=gr.File()
|
| 169 |
)
|
| 170 |
|
|
|
|
|
|
|
|
|
|
| 171 |
# Ejecutar la aplicaci贸n Gradio
|
| 172 |
demo.launch(share=True)
|
|
|
|
| 14 |
|
| 15 |
login(token=os.getenv('HUGGINGFACE_TOKEN'))
|
| 16 |
|
| 17 |
+
# Configuraci贸n del modelo LLM
|
| 18 |
llm = HuggingFaceEndpoint(
|
| 19 |
repo_id="mistralai/Mistral-7B-Instruct-v0.3",
|
| 20 |
task="text-generation",
|
|
|
|
| 37 |
text += pdf_reader.pages[page].extract_text()
|
| 38 |
return text
|
| 39 |
|
| 40 |
+
def summarize(file, summary_length):
|
| 41 |
# Leer el contenido del archivo subido
|
| 42 |
file_path = file.name
|
| 43 |
if file_path.endswith('.pdf'):
|
|
|
|
| 45 |
else:
|
| 46 |
with open(file_path, 'r', encoding='utf-8') as f:
|
| 47 |
text = f.read()
|
| 48 |
+
|
| 49 |
+
if summary_length == 'Corto':
|
| 50 |
+
length_instruction = "El resumen debe tener un m谩ximo de 5 puntos."
|
| 51 |
+
elif summary_length == 'Medio':
|
| 52 |
+
length_instruction = "El resumen debe tener un m谩ximo de 10 puntos."
|
| 53 |
+
else:
|
| 54 |
+
length_instruction = "El resumen debe tener un m谩ximo de 15 puntos."
|
| 55 |
+
|
| 56 |
+
template = f'''
|
| 57 |
+
Por favor, lea detenidamente el siguiente documento:
|
| 58 |
<document>
|
| 59 |
+
{{TEXT}}
|
| 60 |
</document>
|
| 61 |
+
Despu茅s de leer el documento, identifique los puntos clave y las ideas principales cubiertas en el texto. Organice estos puntos clave en una lista con vi帽etas concisa que resuma la informaci贸n esencial del documento. {length_instruction}
|
| 62 |
+
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.
|
| 63 |
'''
|
| 64 |
|
| 65 |
prompt = PromptTemplate(
|
|
|
|
| 70 |
formatted_prompt = prompt.format(TEXT=text)
|
| 71 |
output_summary = llm_engine_hf.invoke(formatted_prompt)
|
| 72 |
|
| 73 |
+
return f"Prompt:\n{formatted_prompt}\n\nResumen:\n{output_summary.content}"
|
| 74 |
|
| 75 |
def classify_text(text):
|
| 76 |
inputs = tokenizer(text, return_tensors="pt", max_length=4096, truncation=True, padding="max_length")
|
|
|
|
| 92 |
text = f.read()
|
| 93 |
|
| 94 |
template = '''
|
| 95 |
+
Por favor, traduzca el siguiente documento al {LANGUAGE}:
|
| 96 |
<document>
|
| 97 |
{TEXT}
|
| 98 |
</document>
|
| 99 |
+
Aseg煤rese de que la traducci贸n sea precisa y conserve el significado original del documento.
|
| 100 |
'''
|
| 101 |
|
| 102 |
prompt = PromptTemplate(
|
|
|
|
| 107 |
formatted_prompt = prompt.format(TEXT=text, LANGUAGE=target_language)
|
| 108 |
translated_text = llm_engine_hf.invoke(formatted_prompt)
|
| 109 |
|
| 110 |
+
return f"Prompt:\n{formatted_prompt}\n\nTraducci贸n:\n{translated_text.content}"
|
| 111 |
|
| 112 |
+
def process_file(file, action, target_language=None, summary_length=None):
|
| 113 |
if action == "Resumen":
|
| 114 |
+
return summarize(file, summary_length)
|
| 115 |
elif action == "Clasificar":
|
| 116 |
file_path = file.name
|
| 117 |
if file_path.endswith('.pdf'):
|
|
|
|
| 125 |
else:
|
| 126 |
return "Acci贸n no v谩lida"
|
| 127 |
|
| 128 |
+
def answer_question(text, question):
|
| 129 |
+
messages = [
|
| 130 |
+
{"role": "system", "content": "Eres un asistente 煤til."},
|
| 131 |
+
{"role": "user", "content": f"El documento es el siguiente:\n{text}"},
|
| 132 |
+
{"role": "user", "content": question}
|
| 133 |
+
]
|
| 134 |
+
response = llm_engine_hf.invoke(messages)
|
| 135 |
+
return response.content
|
| 136 |
+
|
| 137 |
def download_text(output_text, filename='output.txt'):
|
| 138 |
if output_text:
|
| 139 |
file_path = Path(filename)
|
|
|
|
| 149 |
|
| 150 |
# Crear la interfaz de Gradio
|
| 151 |
with gr.Blocks() as demo:
|
| 152 |
+
gr.Markdown("## Procesador de Documentos")
|
| 153 |
|
| 154 |
with gr.Row():
|
| 155 |
with gr.Column():
|
| 156 |
file = gr.File(label="Subir un archivo")
|
| 157 |
action = gr.Radio(label="Seleccione una acci贸n", choices=["Resumen", "Clasificar", "Traducir"])
|
| 158 |
target_language = gr.Dropdown(label="Seleccionar idioma de traducci贸n", choices=["en", "fr", "de"], visible=False)
|
| 159 |
+
summary_length = gr.Radio(label="Seleccione la longitud del resumen", choices=["Corto", "Medio", "Largo"], visible=False)
|
| 160 |
|
| 161 |
with gr.Column():
|
| 162 |
output_text = gr.Textbox(label="Resultado", lines=20)
|
| 163 |
+
question = gr.Textbox(label="Hacer una pregunta al documento", lines=2, visible=False)
|
| 164 |
+
answer = gr.Textbox(label="Respuesta", lines=2, interactive=False, visible=False)
|
| 165 |
|
| 166 |
+
def update_visible_elements(action):
|
| 167 |
if action == "Traducir":
|
| 168 |
+
return gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
|
| 169 |
+
elif action == "Resumen":
|
| 170 |
+
return gr.update(visible=False), gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
|
| 171 |
+
elif action == "Clasificar":
|
| 172 |
+
return gr.update(visible=False), gr.update(visible=False), gr.update(visible=True), gr.update(visible=True), gr.update(visible=False)
|
| 173 |
else:
|
| 174 |
+
return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), gr.update(visible(False)), gr.update(visible=False)
|
| 175 |
|
| 176 |
+
action.change(update_visible_elements, inputs=action, outputs=[target_language, summary_length, question, output_text, answer])
|
| 177 |
|
| 178 |
submit_button = gr.Button("Procesar")
|
| 179 |
+
submit_button.click(process_file, inputs=[file, action, target_language, summary_length], outputs=output_text)
|
| 180 |
|
| 181 |
def generate_file():
|
| 182 |
summary_text = output_text.value
|
|
|
|
| 191 |
outputs=gr.File()
|
| 192 |
)
|
| 193 |
|
| 194 |
+
question_button = gr.Button("Hacer Pregunta")
|
| 195 |
+
question_button.click(answer_question, inputs=[output_text, question], outputs=answer)
|
| 196 |
+
|
| 197 |
# Ejecutar la aplicaci贸n Gradio
|
| 198 |
demo.launch(share=True)
|