Update app.py
Browse files
app.py
CHANGED
|
@@ -1,64 +1,144 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
-
from transformers import
|
| 3 |
-
from transformers import T5Tokenizer # Import T5Tokenizer directly
|
| 4 |
import torch
|
| 5 |
|
| 6 |
-
#
|
| 7 |
-
|
| 8 |
-
# Use T5Tokenizer directly instead of AutoTokenizer
|
| 9 |
-
simplifier_tokenizer = T5Tokenizer.from_pretrained(simplifier_model_name)
|
| 10 |
-
simplifier_model = AutoModelForSeq2SeqLM.from_pretrained(simplifier_model_name)
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
}
|
| 18 |
-
prompt = f"{niveles[nivel]}\n\n{texto}"
|
| 19 |
-
inputs = simplifier_tokenizer(prompt, return_tensors="pt", truncation=True)
|
| 20 |
-
outputs = simplifier_model.generate(
|
| 21 |
-
**inputs,
|
| 22 |
-
max_new_tokens=120,
|
| 23 |
-
num_beams=4,
|
| 24 |
-
temperature=0.7,
|
| 25 |
-
repetition_penalty=1.2,
|
| 26 |
-
early_stopping=True
|
| 27 |
-
)
|
| 28 |
-
resultado = simplifier_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 29 |
-
return resultado
|
| 30 |
|
| 31 |
-
#
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
predictor_model = AutoModelForCausalLM.from_pretrained(predictor_model_name)
|
| 35 |
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
texto_generado = predictor_tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 40 |
-
return texto_generado[len(texto_inicial):] # Solo mostrar lo nuevo
|
| 41 |
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
with gr.
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
|
| 64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
from transformers import GPT2LMHeadModel, GPT2Tokenizer
|
|
|
|
| 3 |
import torch
|
| 4 |
|
| 5 |
+
# Configurar el dispositivo (CPU)
|
| 6 |
+
device = torch.device("cpu")
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
+
# Cargar el modelo y tokenizer
|
| 9 |
+
print("Cargando modelo DistilGPT-2...")
|
| 10 |
+
model_name = "distilgpt2"
|
| 11 |
+
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
|
| 12 |
+
model = GPT2LMHeadModel.from_pretrained(model_name)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
+
# Mover modelo a CPU y ponerlo en modo evaluaci贸n
|
| 15 |
+
model.to(device)
|
| 16 |
+
model.eval()
|
|
|
|
| 17 |
|
| 18 |
+
# Configurar pad_token si no existe
|
| 19 |
+
if tokenizer.pad_token is None:
|
| 20 |
+
tokenizer.pad_token = tokenizer.eos_token
|
|
|
|
|
|
|
| 21 |
|
| 22 |
+
def autocomplete_text(input_text, max_tokens=20):
|
| 23 |
+
"""
|
| 24 |
+
Autocompleta el texto de entrada usando DistilGPT-2
|
| 25 |
+
|
| 26 |
+
Args:
|
| 27 |
+
input_text (str): Texto inicial a completar
|
| 28 |
+
max_tokens (int): N煤mero m谩ximo de tokens a generar
|
| 29 |
+
|
| 30 |
+
Returns:
|
| 31 |
+
str: Solo la parte nueva generada (sin el input original)
|
| 32 |
+
"""
|
| 33 |
+
if not input_text.strip():
|
| 34 |
+
return "Por favor, ingresa alg煤n texto para completar."
|
| 35 |
+
|
| 36 |
+
try:
|
| 37 |
+
# Tokenizar el texto de entrada
|
| 38 |
+
inputs = tokenizer.encode(input_text, return_tensors="pt", padding=True)
|
| 39 |
+
inputs = inputs.to(device)
|
| 40 |
+
|
| 41 |
+
# Generar texto
|
| 42 |
+
with torch.no_grad():
|
| 43 |
+
outputs = model.generate(
|
| 44 |
+
inputs,
|
| 45 |
+
max_new_tokens=max_tokens,
|
| 46 |
+
num_return_sequences=1,
|
| 47 |
+
temperature=0.7,
|
| 48 |
+
do_sample=True,
|
| 49 |
+
pad_token_id=tokenizer.eos_token_id,
|
| 50 |
+
eos_token_id=tokenizer.eos_token_id,
|
| 51 |
+
attention_mask=torch.ones_like(inputs)
|
| 52 |
+
)
|
| 53 |
+
|
| 54 |
+
# Decodificar el resultado completo
|
| 55 |
+
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 56 |
+
|
| 57 |
+
# Extraer solo la parte nueva (sin el input original)
|
| 58 |
+
new_text = generated_text[len(input_text):].strip()
|
| 59 |
+
|
| 60 |
+
if not new_text:
|
| 61 |
+
return "No se pudo generar texto adicional."
|
| 62 |
+
|
| 63 |
+
return new_text
|
| 64 |
+
|
| 65 |
+
except Exception as e:
|
| 66 |
+
return f"Error al generar texto: {str(e)}"
|
| 67 |
|
| 68 |
+
def create_autocomplete_interface():
|
| 69 |
+
"""
|
| 70 |
+
Crea la interfaz de autocompletar dentro de gr.Blocks()
|
| 71 |
+
"""
|
| 72 |
+
|
| 73 |
+
with gr.Blocks(title="Autocompletar Texto") as demo:
|
| 74 |
+
|
| 75 |
+
gr.Markdown("# 馃 Autocompletar Texto")
|
| 76 |
+
gr.Markdown("Escribe el inicio de una frase y la IA la completar谩 por ti.")
|
| 77 |
+
|
| 78 |
+
with gr.Tab("Autocompletar"):
|
| 79 |
+
with gr.Row():
|
| 80 |
+
with gr.Column():
|
| 81 |
+
input_textbox = gr.Textbox(
|
| 82 |
+
label="Texto a completar",
|
| 83 |
+
placeholder="Escribe el inicio de tu frase aqu铆...",
|
| 84 |
+
lines=3,
|
| 85 |
+
max_lines=5
|
| 86 |
+
)
|
| 87 |
+
|
| 88 |
+
generate_btn = gr.Button("Completar Texto", variant="primary")
|
| 89 |
+
|
| 90 |
+
with gr.Column():
|
| 91 |
+
output_textbox = gr.Textbox(
|
| 92 |
+
label="Texto generado",
|
| 93 |
+
placeholder="Aqu铆 aparecer谩 la continuaci贸n...",
|
| 94 |
+
lines=3,
|
| 95 |
+
max_lines=5,
|
| 96 |
+
interactive=False
|
| 97 |
+
)
|
| 98 |
+
|
| 99 |
+
# Conectar el bot贸n con la funci贸n
|
| 100 |
+
generate_btn.click(
|
| 101 |
+
fn=autocomplete_text,
|
| 102 |
+
inputs=[input_textbox],
|
| 103 |
+
outputs=[output_textbox]
|
| 104 |
+
)
|
| 105 |
+
|
| 106 |
+
# Tambi茅n permitir Enter para generar
|
| 107 |
+
input_textbox.submit(
|
| 108 |
+
fn=autocomplete_text,
|
| 109 |
+
inputs=[input_textbox],
|
| 110 |
+
outputs=[output_textbox]
|
| 111 |
+
)
|
| 112 |
+
|
| 113 |
+
# Pesta帽a adicional con ejemplos
|
| 114 |
+
with gr.Tab("Ejemplos"):
|
| 115 |
+
gr.Markdown("""
|
| 116 |
+
### Ejemplos de uso:
|
| 117 |
+
|
| 118 |
+
**Entrada:** "El clima de hoy est谩"
|
| 119 |
+
**Salida:** "muy agradable y soleado"
|
| 120 |
+
|
| 121 |
+
**Entrada:** "Me gusta mucho"
|
| 122 |
+
**Salida:** "pasar tiempo con mi familia"
|
| 123 |
+
|
| 124 |
+
**Entrada:** "Para hacer una buena comida necesitas"
|
| 125 |
+
**Salida:** "ingredientes frescos y mucha paciencia"
|
| 126 |
+
""")
|
| 127 |
+
|
| 128 |
+
return demo
|
| 129 |
|
| 130 |
+
# Crear y lanzar la aplicaci贸n
|
| 131 |
+
if __name__ == "__main__":
|
| 132 |
+
print("Iniciando aplicaci贸n de autocompletar...")
|
| 133 |
+
|
| 134 |
+
# Crear la interfaz
|
| 135 |
+
app = create_autocomplete_interface()
|
| 136 |
+
|
| 137 |
+
# Lanzar la aplicaci贸n
|
| 138 |
+
app.launch(
|
| 139 |
+
share=False, # Cambiar a True si quieres compartir p煤blicamente
|
| 140 |
+
server_name="0.0.0.0", # Permite acceso desde otras m谩quinas en la red local
|
| 141 |
+
server_port=7860, # Puerto por defecto de Gradio
|
| 142 |
+
show_error=True,
|
| 143 |
+
debug=False
|
| 144 |
+
)
|