Spaces:
Paused
Paused
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,70 +1,26 @@
|
|
| 1 |
-
# chat_ai.py
|
| 2 |
-
|
| 3 |
-
# ruff: noqa: E402
|
| 4 |
-
# Above allows ruff to ignore E402: module level import not at top of file
|
| 5 |
-
|
| 6 |
-
import re
|
| 7 |
-
import tempfile
|
| 8 |
-
import os
|
| 9 |
-
|
| 10 |
import torch
|
| 11 |
-
import
|
| 12 |
import gradio as gr
|
| 13 |
-
import numpy as np
|
| 14 |
-
import soundfile as sf
|
| 15 |
-
import torchaudio
|
| 16 |
-
from cached_path import cached_path
|
| 17 |
-
from transformers import (
|
| 18 |
-
AutoModelForCausalLM,
|
| 19 |
-
AutoTokenizer,
|
| 20 |
-
WhisperProcessor,
|
| 21 |
-
WhisperForConditionalGeneration,
|
| 22 |
-
)
|
| 23 |
-
from num2words import num2words
|
| 24 |
-
|
| 25 |
-
try:
|
| 26 |
-
import spaces
|
| 27 |
-
USING_SPACES = True
|
| 28 |
-
except ImportError:
|
| 29 |
-
USING_SPACES = False
|
| 30 |
-
|
| 31 |
-
def gpu_decorator(func):
|
| 32 |
-
if USING_SPACES:
|
| 33 |
-
return spaces.GPU(func)
|
| 34 |
-
else:
|
| 35 |
-
return func
|
| 36 |
-
|
| 37 |
-
from f5_tts.model import DiT
|
| 38 |
-
from f5_tts.infer.utils_infer import (
|
| 39 |
-
load_vocoder,
|
| 40 |
-
load_model,
|
| 41 |
-
preprocess_ref_audio_text,
|
| 42 |
-
infer_process,
|
| 43 |
-
remove_silence_for_generated_wav,
|
| 44 |
-
save_spectrogram,
|
| 45 |
-
)
|
| 46 |
-
|
| 47 |
-
# Cargar el vocoder
|
| 48 |
-
vocoder = load_vocoder()
|
| 49 |
-
|
| 50 |
-
# Configuración y carga del modelo F5-TTS
|
| 51 |
-
F5TTS_model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)
|
| 52 |
-
F5TTS_ema_model = load_model(
|
| 53 |
-
DiT, F5TTS_model_cfg, str(cached_path("hf://jpgallegoar/F5-Spanish/model_1200000.safetensors"))
|
| 54 |
-
)
|
| 55 |
|
| 56 |
# Variables globales para el modelo de chat
|
| 57 |
chat_model_state = None
|
| 58 |
chat_tokenizer_state = None
|
| 59 |
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
|
| 67 |
-
@gpu_decorator
|
| 68 |
def generate_response(messages, model, tokenizer):
|
| 69 |
"""Genera una respuesta usando el modelo de chat"""
|
| 70 |
try:
|
|
@@ -100,397 +56,68 @@ def generate_response(messages, model, tokenizer):
|
|
| 100 |
|
| 101 |
# Extraer solo la respuesta del asistente
|
| 102 |
response = generated_text[len(prompt):].strip()
|
| 103 |
-
# Opcional: Cortar la respuesta al primer salto de línea
|
| 104 |
-
response = response.split("\n")[0]
|
| 105 |
return response
|
| 106 |
except Exception as e:
|
| 107 |
-
# Log del error para depuración
|
| 108 |
print(f"Error en generate_response: {e}")
|
| 109 |
return "Lo siento, ocurrió un error al generar la respuesta."
|
| 110 |
|
| 111 |
-
|
| 112 |
-
"""Convierte números en texto a su representación en palabras en español"""
|
| 113 |
-
texto_separado = re.sub(r'([A-Za-z])(\d)', r'\1 \2', texto)
|
| 114 |
-
texto_separado = re.sub(r'(\d)([A-Za-z])', r'\1 \2', texto_separado)
|
| 115 |
-
|
| 116 |
-
def reemplazar_numero(match):
|
| 117 |
-
numero = match.group()
|
| 118 |
-
try:
|
| 119 |
-
return num2words(int(numero), lang='es')
|
| 120 |
-
except ValueError:
|
| 121 |
-
return numero
|
| 122 |
-
|
| 123 |
-
texto_traducido = re.sub(r'\b\d+\b', reemplazar_numero, texto_separado)
|
| 124 |
-
|
| 125 |
-
return texto_traducido
|
| 126 |
-
|
| 127 |
-
@gpu_decorator
|
| 128 |
-
def infer(
|
| 129 |
-
ref_audio_orig, ref_text, gen_text, model, remove_silence, cross_fade_duration=0.15, speed=1
|
| 130 |
-
):
|
| 131 |
-
"""Genera el audio sintetizado a partir del texto"""
|
| 132 |
-
try:
|
| 133 |
-
ref_audio, ref_text = preprocess_ref_audio_text(ref_audio_orig, ref_text)
|
| 134 |
-
|
| 135 |
-
ema_model = F5TTS_ema_model
|
| 136 |
-
|
| 137 |
-
if not gen_text.startswith(" "):
|
| 138 |
-
gen_text = " " + gen_text
|
| 139 |
-
if not gen_text.endswith(". "):
|
| 140 |
-
gen_text += ". "
|
| 141 |
-
|
| 142 |
-
gen_text = gen_text.lower()
|
| 143 |
-
gen_text = traducir_numero_a_texto(gen_text)
|
| 144 |
-
|
| 145 |
-
final_wave, final_sample_rate, combined_spectrogram = infer_process(
|
| 146 |
-
ref_audio,
|
| 147 |
-
ref_text,
|
| 148 |
-
gen_text,
|
| 149 |
-
ema_model,
|
| 150 |
-
vocoder,
|
| 151 |
-
cross_fade_duration=cross_fade_duration,
|
| 152 |
-
speed=speed,
|
| 153 |
-
progress=gr.Progress(),
|
| 154 |
-
)
|
| 155 |
-
|
| 156 |
-
# Eliminar silencios si está activado
|
| 157 |
-
if remove_silence:
|
| 158 |
-
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
|
| 159 |
-
sf.write(f.name, final_wave, final_sample_rate)
|
| 160 |
-
remove_silence_for_generated_wav(f.name)
|
| 161 |
-
final_wave, _ = torchaudio.load(f.name)
|
| 162 |
-
final_wave = final_wave.squeeze().cpu().numpy()
|
| 163 |
-
|
| 164 |
-
# Guardar el espectrograma (opcional)
|
| 165 |
-
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as tmp_spectrogram:
|
| 166 |
-
spectrogram_path = tmp_spectrogram.name
|
| 167 |
-
save_spectrogram(combined_spectrogram, spectrogram_path)
|
| 168 |
-
|
| 169 |
-
return (final_sample_rate, final_wave), spectrogram_path
|
| 170 |
-
except Exception as e:
|
| 171 |
-
# Log del error para depuración
|
| 172 |
-
print(f"Error en infer: {e}")
|
| 173 |
-
return None, None
|
| 174 |
-
|
| 175 |
-
def load_chat_model_function():
|
| 176 |
-
"""Función para cargar el modelo de chat"""
|
| 177 |
-
global chat_model_state, chat_tokenizer_state
|
| 178 |
-
if chat_model_state is None:
|
| 179 |
-
try:
|
| 180 |
-
model_name = "Qwen/Qwen2.5-3B-Instruct"
|
| 181 |
-
chat_model_state = AutoModelForCausalLM.from_pretrained(
|
| 182 |
-
model_name, torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32, device_map="auto" if torch.cuda.is_available() else None
|
| 183 |
-
)
|
| 184 |
-
chat_tokenizer_state = AutoTokenizer.from_pretrained(model_name)
|
| 185 |
-
return gr.update(visible=False), gr.update(visible=True)
|
| 186 |
-
except Exception as e:
|
| 187 |
-
print(f"Error al cargar el modelo de chat: {e}")
|
| 188 |
-
return gr.update(value="Error al cargar el modelo de chat."), gr.update(visible=False)
|
| 189 |
-
else:
|
| 190 |
-
return gr.update(visible=False), gr.update(visible=True)
|
| 191 |
-
|
| 192 |
-
def transcribe_audio(audio_path):
|
| 193 |
-
"""Transcribe el audio usando el modelo Whisper"""
|
| 194 |
-
try:
|
| 195 |
-
if not os.path.exists(audio_path):
|
| 196 |
-
raise FileNotFoundError(f"Archivo de audio no encontrado: {audio_path}")
|
| 197 |
-
|
| 198 |
-
# Cargar el audio
|
| 199 |
-
audio, rate = torchaudio.load(audio_path)
|
| 200 |
-
|
| 201 |
-
# Resample si es necesario
|
| 202 |
-
if rate != 16000:
|
| 203 |
-
resampler = torchaudio.transforms.Resample(orig_freq=rate, new_freq=16000)
|
| 204 |
-
audio = resampler(audio)
|
| 205 |
-
|
| 206 |
-
# Asegurarse de que el audio tenga una sola dimensión
|
| 207 |
-
if audio.ndim > 1:
|
| 208 |
-
audio = torch.mean(audio, dim=0)
|
| 209 |
-
|
| 210 |
-
input_features = whisper_processor(audio.cpu().numpy(), sampling_rate=16000, return_tensors="pt").input_features
|
| 211 |
-
if torch.cuda.is_available():
|
| 212 |
-
input_features = input_features.to("cuda")
|
| 213 |
-
|
| 214 |
-
# Generar la transcripción
|
| 215 |
-
predicted_ids = whisper_model.generate(input_features)
|
| 216 |
-
transcription = whisper_processor.decode(predicted_ids[0], skip_special_tokens=True)
|
| 217 |
-
|
| 218 |
-
return transcription
|
| 219 |
-
except Exception as e:
|
| 220 |
-
print(f"Error en transcribe_audio: {e}")
|
| 221 |
-
return None
|
| 222 |
-
|
| 223 |
with gr.Blocks() as app_chat:
|
| 224 |
-
gr.Markdown(
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
|
| 230 |
-
|
| 231 |
-
|
| 232 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 233 |
)
|
| 234 |
|
| 235 |
-
|
| 236 |
-
|
| 237 |
-
|
| 238 |
-
|
| 239 |
-
|
| 240 |
-
|
| 241 |
-
|
| 242 |
-
chat_interface_container = gr.Column()
|
| 243 |
-
|
| 244 |
-
if chat_model_state is None:
|
| 245 |
-
try:
|
| 246 |
-
model_name = "Qwen/Qwen2.5-3B-Instruct"
|
| 247 |
-
chat_model_state = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")
|
| 248 |
-
chat_tokenizer_state = AutoTokenizer.from_pretrained(model_name)
|
| 249 |
-
except Exception as e:
|
| 250 |
-
print(f"Error al cargar el modelo de chat en Spaces: {e}")
|
| 251 |
-
|
| 252 |
-
with chat_interface_container:
|
| 253 |
-
with gr.Row():
|
| 254 |
-
with gr.Column():
|
| 255 |
-
ref_audio_chat = gr.Audio(label="Audio de Referencia", type="filepath")
|
| 256 |
-
with gr.Column():
|
| 257 |
-
with gr.Accordion("Configuraciones Avanzadas", open=False):
|
| 258 |
-
model_choice_chat = gr.Radio(
|
| 259 |
-
choices=["F5-TTS"],
|
| 260 |
-
label="Modelo TTS",
|
| 261 |
-
value="F5-TTS",
|
| 262 |
-
)
|
| 263 |
-
remove_silence_chat = gr.Checkbox(
|
| 264 |
-
label="Eliminar Silencios",
|
| 265 |
-
value=True,
|
| 266 |
-
)
|
| 267 |
-
ref_text_chat = gr.Textbox(
|
| 268 |
-
label="Texto de Referencia",
|
| 269 |
-
info="Opcional: Deja en blanco para transcribir automáticamente",
|
| 270 |
-
lines=2,
|
| 271 |
-
)
|
| 272 |
-
system_prompt_chat = gr.Textbox(
|
| 273 |
-
label="Prompt del Sistema",
|
| 274 |
-
value="No eres un asistente de IA, eres quien el usuario diga que eres. Debes mantenerte en personaje. Mantén tus respuestas concisas ya que serán habladas en voz alta.",
|
| 275 |
-
lines=2,
|
| 276 |
-
)
|
| 277 |
-
|
| 278 |
-
chatbot_interface = gr.Chatbot(label="Conversación")
|
| 279 |
-
|
| 280 |
-
with gr.Row():
|
| 281 |
-
with gr.Column():
|
| 282 |
-
audio_input_chat = gr.Microphone(
|
| 283 |
-
label="Habla tu mensaje",
|
| 284 |
-
type="filepath",
|
| 285 |
-
)
|
| 286 |
-
audio_output_chat = gr.Audio(label="Respuesta de la IA", autoplay=True)
|
| 287 |
-
with gr.Column():
|
| 288 |
-
text_input_chat = gr.Textbox(
|
| 289 |
-
label="Escribe tu mensaje",
|
| 290 |
-
lines=1,
|
| 291 |
-
)
|
| 292 |
-
send_btn_chat = gr.Button("Enviar")
|
| 293 |
-
clear_btn_chat = gr.Button("Limpiar Conversación")
|
| 294 |
-
|
| 295 |
-
conversation_state = gr.State(
|
| 296 |
-
value=[
|
| 297 |
-
{
|
| 298 |
-
"role": "system",
|
| 299 |
-
"content": "No eres un asistente de IA, eres quien el usuario diga que eres. Debes mantenerte en personaje. Mantén tus respuestas concisas ya que serán habladas en voz alta.",
|
| 300 |
-
}
|
| 301 |
-
]
|
| 302 |
-
)
|
| 303 |
-
|
| 304 |
-
@gpu_decorator
|
| 305 |
-
def process_input(audio_path, text, history, conv_state, ref_text):
|
| 306 |
-
"""Procesa la entrada de audio o texto del usuario y genera una respuesta."""
|
| 307 |
-
try:
|
| 308 |
-
if not audio_path and not text.strip():
|
| 309 |
-
return history, conv_state, ""
|
| 310 |
-
|
| 311 |
-
if audio_path:
|
| 312 |
-
# Transcribir el audio usando Whisper
|
| 313 |
-
transcribed_text = transcribe_audio(audio_path)
|
| 314 |
-
if transcribed_text is None:
|
| 315 |
-
history.append(("Error en la transcripción de audio.", None))
|
| 316 |
-
return history, conv_state, "Lo siento, ocurrió un error al procesar tu audio."
|
| 317 |
-
|
| 318 |
-
text = transcribed_text
|
| 319 |
-
|
| 320 |
-
if not text.strip():
|
| 321 |
-
return history, conv_state, ""
|
| 322 |
-
|
| 323 |
-
# Si se proporciona texto de referencia, usarlo; de lo contrario, usar transcripción
|
| 324 |
-
if ref_text.strip():
|
| 325 |
-
input_text = ref_text + " " + text
|
| 326 |
-
else:
|
| 327 |
-
input_text = text
|
| 328 |
-
|
| 329 |
-
conv_state.append({"role": "user", "content": input_text})
|
| 330 |
-
history.append((input_text, None))
|
| 331 |
-
|
| 332 |
-
# Generar la respuesta del modelo de chat
|
| 333 |
-
response = generate_response(conv_state, chat_model_state, chat_tokenizer_state)
|
| 334 |
-
|
| 335 |
-
conv_state.append({"role": "assistant", "content": response})
|
| 336 |
-
history[-1] = (input_text, response)
|
| 337 |
-
|
| 338 |
-
return history, conv_state, response
|
| 339 |
-
except Exception as e:
|
| 340 |
-
print(f"Error en process_input: {e}")
|
| 341 |
-
history.append(("Error al procesar tu solicitud.", None))
|
| 342 |
-
return history, conv_state, "Lo siento, ocurrió un error al procesar tu solicitud."
|
| 343 |
-
|
| 344 |
-
@gpu_decorator
|
| 345 |
-
def generate_audio_response(response, ref_audio, ref_text, model, remove_silence):
|
| 346 |
-
"""Genera el audio de respuesta para la IA."""
|
| 347 |
-
try:
|
| 348 |
-
if not response or not ref_audio:
|
| 349 |
-
return None
|
| 350 |
-
|
| 351 |
-
# Verificar si la respuesta es un mensaje de error
|
| 352 |
-
if response.startswith("Lo siento"):
|
| 353 |
-
# Aquí podrías tener un audio pregrabado de error
|
| 354 |
-
# Por ejemplo, "error_audio.wav" en el mismo directorio
|
| 355 |
-
error_audio_path = "error_audio.wav"
|
| 356 |
-
if os.path.exists(error_audio_path):
|
| 357 |
-
return error_audio_path
|
| 358 |
-
else:
|
| 359 |
-
# Si no tienes un archivo de audio de error, puedes retornar None
|
| 360 |
-
# O generar el audio dinámicamente usando infer
|
| 361 |
-
audio_result, _ = infer(
|
| 362 |
-
ref_audio,
|
| 363 |
-
ref_text,
|
| 364 |
-
response,
|
| 365 |
-
model,
|
| 366 |
-
remove_silence,
|
| 367 |
-
cross_fade_duration=0.15,
|
| 368 |
-
speed=1.0,
|
| 369 |
-
)
|
| 370 |
-
return audio_result
|
| 371 |
-
else:
|
| 372 |
-
# Generar el audio de la respuesta normal
|
| 373 |
-
audio_result, _ = infer(
|
| 374 |
-
ref_audio,
|
| 375 |
-
ref_text,
|
| 376 |
-
response,
|
| 377 |
-
model,
|
| 378 |
-
remove_silence,
|
| 379 |
-
cross_fade_duration=0.15,
|
| 380 |
-
speed=1.0,
|
| 381 |
-
)
|
| 382 |
-
if audio_result is None:
|
| 383 |
-
# Retornar un audio de error si infer falla
|
| 384 |
-
error_audio_path = "error_audio.wav"
|
| 385 |
-
if os.path.exists(error_audio_path):
|
| 386 |
-
return error_audio_path
|
| 387 |
-
else:
|
| 388 |
-
return None
|
| 389 |
-
return audio_result
|
| 390 |
-
except Exception as e:
|
| 391 |
-
print(f"Error en generate_audio_response: {e}")
|
| 392 |
-
# Retornar un audio de error si ocurre una excepción
|
| 393 |
-
error_audio_path = "error_audio.wav"
|
| 394 |
-
if os.path.exists(error_audio_path):
|
| 395 |
-
return error_audio_path
|
| 396 |
-
else:
|
| 397 |
-
return None
|
| 398 |
|
| 399 |
-
|
| 400 |
-
|
| 401 |
-
return [], [
|
| 402 |
-
{
|
| 403 |
-
"role": "system",
|
| 404 |
-
"content": "No eres un asistente de IA, eres quien el usuario diga que eres. Debes mantenerte en personaje. Mantén tus respuestas concisas ya que serán habladas en voz alta.",
|
| 405 |
-
}
|
| 406 |
-
]
|
| 407 |
|
| 408 |
-
|
| 409 |
-
|
| 410 |
-
new_conv_state = [{"role": "system", "content": new_prompt}]
|
| 411 |
-
return [], new_conv_state
|
| 412 |
|
| 413 |
-
|
| 414 |
-
audio_input_chat.change(
|
| 415 |
-
process_input,
|
| 416 |
-
inputs=[audio_input_chat, text_input_chat, chatbot_interface, conversation_state, ref_text_chat],
|
| 417 |
-
outputs=[chatbot_interface, conversation_state, text_input_chat],
|
| 418 |
-
).then(
|
| 419 |
-
generate_audio_response,
|
| 420 |
-
inputs=[chatbot_interface, ref_audio_chat, ref_text_chat, model_choice_chat, remove_silence_chat],
|
| 421 |
-
outputs=[audio_output_chat],
|
| 422 |
-
).then(
|
| 423 |
-
lambda: None,
|
| 424 |
-
None,
|
| 425 |
-
audio_input_chat,
|
| 426 |
-
)
|
| 427 |
|
| 428 |
-
|
| 429 |
-
|
| 430 |
-
|
| 431 |
-
inputs=[audio_input_chat, text_input_chat, chatbot_interface, conversation_state, ref_text_chat],
|
| 432 |
-
outputs=[chatbot_interface, conversation_state, text_input_chat],
|
| 433 |
-
).then(
|
| 434 |
-
generate_audio_response,
|
| 435 |
-
inputs=[chatbot_interface, ref_audio_chat, ref_text_chat, model_choice_chat, remove_silence_chat],
|
| 436 |
-
outputs=[audio_output_chat],
|
| 437 |
-
).then(
|
| 438 |
-
lambda: None,
|
| 439 |
-
None,
|
| 440 |
-
text_input_chat,
|
| 441 |
-
)
|
| 442 |
|
| 443 |
-
|
| 444 |
-
|
| 445 |
-
|
| 446 |
-
|
| 447 |
-
|
| 448 |
-
|
| 449 |
-
generate_audio_response,
|
| 450 |
-
inputs=[chatbot_interface, ref_audio_chat, ref_text_chat, model_choice_chat, remove_silence_chat],
|
| 451 |
-
outputs=[audio_output_chat],
|
| 452 |
-
).then(
|
| 453 |
-
lambda: None,
|
| 454 |
-
None,
|
| 455 |
-
text_input_chat,
|
| 456 |
-
)
|
| 457 |
-
|
| 458 |
-
# Manejar el botón de limpiar conversación
|
| 459 |
-
clear_btn_chat.click(
|
| 460 |
-
clear_conversation,
|
| 461 |
-
outputs=[chatbot_interface, conversation_state],
|
| 462 |
-
)
|
| 463 |
|
| 464 |
-
|
| 465 |
-
|
| 466 |
-
|
| 467 |
-
|
| 468 |
-
|
| 469 |
-
)
|
| 470 |
|
| 471 |
-
|
| 472 |
-
|
| 473 |
-
|
| 474 |
-
@click.option(
|
| 475 |
-
"--share",
|
| 476 |
-
"-s",
|
| 477 |
-
default=False,
|
| 478 |
-
is_flag=True,
|
| 479 |
-
help="Compartir la aplicación a través de un enlace compartido de Gradio",
|
| 480 |
-
)
|
| 481 |
-
@click.option("--api", "-a", default=True, is_flag=True, help="Permitir acceso a la API")
|
| 482 |
-
def main(port, host, share, api):
|
| 483 |
-
"""Función principal para lanzar la aplicación Gradio de Chat AI."""
|
| 484 |
-
print("Iniciando la aplicación de Chat AI...")
|
| 485 |
-
app_chat.queue(api_open=api).launch(
|
| 486 |
-
server_name=host,
|
| 487 |
-
server_port=port,
|
| 488 |
-
share=share,
|
| 489 |
-
show_api=api
|
| 490 |
)
|
| 491 |
|
| 492 |
-
|
| 493 |
-
|
| 494 |
-
|
| 495 |
-
|
| 496 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 3 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
|
| 5 |
# Variables globales para el modelo de chat
|
| 6 |
chat_model_state = None
|
| 7 |
chat_tokenizer_state = None
|
| 8 |
|
| 9 |
+
def load_chat_model():
|
| 10 |
+
"""Función para cargar el modelo de chat"""
|
| 11 |
+
global chat_model_state, chat_tokenizer_state
|
| 12 |
+
try:
|
| 13 |
+
model_name = "Qwen/Qwen2.5-3B-Instruct"
|
| 14 |
+
chat_model_state = AutoModelForCausalLM.from_pretrained(
|
| 15 |
+
model_name,
|
| 16 |
+
torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
|
| 17 |
+
device_map="auto" if torch.cuda.is_available() else None
|
| 18 |
+
)
|
| 19 |
+
chat_tokenizer_state = AutoTokenizer.from_pretrained(model_name)
|
| 20 |
+
print("Modelo cargado exitosamente.")
|
| 21 |
+
except Exception as e:
|
| 22 |
+
print(f"Error al cargar el modelo de chat: {e}")
|
| 23 |
|
|
|
|
| 24 |
def generate_response(messages, model, tokenizer):
|
| 25 |
"""Genera una respuesta usando el modelo de chat"""
|
| 26 |
try:
|
|
|
|
| 56 |
|
| 57 |
# Extraer solo la respuesta del asistente
|
| 58 |
response = generated_text[len(prompt):].strip()
|
|
|
|
|
|
|
| 59 |
return response
|
| 60 |
except Exception as e:
|
|
|
|
| 61 |
print(f"Error en generate_response: {e}")
|
| 62 |
return "Lo siento, ocurrió un error al generar la respuesta."
|
| 63 |
|
| 64 |
+
# Gradio Interface
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
with gr.Blocks() as app_chat:
|
| 66 |
+
gr.Markdown("### Chatbot Simple")
|
| 67 |
+
chatbot_interface = gr.Chatbot(label="Conversación")
|
| 68 |
+
text_input_chat = gr.Textbox(label="Escribe tu mensaje", lines=1)
|
| 69 |
+
send_btn_chat = gr.Button("Enviar")
|
| 70 |
+
clear_btn_chat = gr.Button("Limpiar Conversación")
|
| 71 |
+
|
| 72 |
+
conversation_state = gr.State(
|
| 73 |
+
value=[
|
| 74 |
+
{
|
| 75 |
+
"role": "system",
|
| 76 |
+
"content": "Eres un chatbot. Responde a las preguntas del usuario de manera concisa y clara.",
|
| 77 |
+
}
|
| 78 |
+
]
|
| 79 |
)
|
| 80 |
|
| 81 |
+
def process_input(text, history, conv_state):
|
| 82 |
+
"""Procesa la entrada de texto del usuario y genera una respuesta."""
|
| 83 |
+
if not text.strip():
|
| 84 |
+
return history, conv_state, ""
|
| 85 |
+
|
| 86 |
+
conv_state.append({"role": "user", "content": text})
|
| 87 |
+
history.append((text, None))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 88 |
|
| 89 |
+
# Generar la respuesta del modelo de chat
|
| 90 |
+
response = generate_response(conv_state, chat_model_state, chat_tokenizer_state)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
|
| 92 |
+
conv_state.append({"role": "assistant", "content": response})
|
| 93 |
+
history[-1] = (text, response)
|
|
|
|
|
|
|
| 94 |
|
| 95 |
+
return history, conv_state, ""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 96 |
|
| 97 |
+
def clear_conversation():
|
| 98 |
+
"""Resetea la conversación"""
|
| 99 |
+
return [], [{"role": "system", "content": "Eres un chatbot. Responde a las preguntas del usuario de manera concisa y clara."}]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 100 |
|
| 101 |
+
# Manejar entrada de texto y botones
|
| 102 |
+
text_input_chat.submit(
|
| 103 |
+
process_input,
|
| 104 |
+
inputs=[text_input_chat, chatbot_interface, conversation_state],
|
| 105 |
+
outputs=[chatbot_interface, conversation_state, text_input_chat],
|
| 106 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
|
| 108 |
+
send_btn_chat.click(
|
| 109 |
+
process_input,
|
| 110 |
+
inputs=[text_input_chat, chatbot_interface, conversation_state],
|
| 111 |
+
outputs=[chatbot_interface, conversation_state, text_input_chat],
|
| 112 |
+
)
|
|
|
|
| 113 |
|
| 114 |
+
clear_btn_chat.click(
|
| 115 |
+
clear_conversation,
|
| 116 |
+
outputs=[chatbot_interface, conversation_state],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 117 |
)
|
| 118 |
|
| 119 |
+
# Cargar el modelo al iniciar
|
| 120 |
+
load_chat_model()
|
| 121 |
+
|
| 122 |
+
# Ejecutar la aplicación
|
| 123 |
+
app_chat.launch()
|