Agregando más modelos: chat, traducción y optimizaciones de velocidad
Browse files
app.py
CHANGED
|
@@ -11,13 +11,42 @@ import base64
|
|
| 11 |
MODELS = {
|
| 12 |
"text": {
|
| 13 |
"microsoft/DialoGPT-medium": "Chat conversacional",
|
|
|
|
|
|
|
| 14 |
"gpt2": "Generación de texto",
|
|
|
|
|
|
|
| 15 |
"distilgpt2": "GPT-2 optimizado",
|
| 16 |
-
"EleutherAI/gpt-neo-125M": "GPT-Neo pequeño"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
},
|
| 18 |
"image": {
|
| 19 |
"runwayml/stable-diffusion-v1-5": "Stable Diffusion v1.5",
|
| 20 |
-
"CompVis/stable-diffusion-v1-4": "Stable Diffusion v1.4"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
}
|
| 22 |
}
|
| 23 |
|
|
@@ -25,16 +54,25 @@ MODELS = {
|
|
| 25 |
model_cache = {}
|
| 26 |
|
| 27 |
def load_text_model(model_name):
|
| 28 |
-
"""Cargar modelo de texto"""
|
| 29 |
if model_name not in model_cache:
|
| 30 |
print(f"Cargando modelo de texto: {model_name}")
|
| 31 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 32 |
-
model = AutoModelForCausalLM.from_pretrained(model_name)
|
| 33 |
|
| 34 |
-
#
|
| 35 |
-
if "
|
| 36 |
-
|
| 37 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 38 |
|
| 39 |
model_cache[model_name] = {
|
| 40 |
"tokenizer": tokenizer,
|
|
@@ -45,16 +83,21 @@ def load_text_model(model_name):
|
|
| 45 |
return model_cache[model_name]
|
| 46 |
|
| 47 |
def load_image_model(model_name):
|
| 48 |
-
"""Cargar modelo de imagen"""
|
| 49 |
if model_name not in model_cache:
|
| 50 |
print(f"Cargando modelo de imagen: {model_name}")
|
|
|
|
|
|
|
| 51 |
pipe = StableDiffusionPipeline.from_pretrained(
|
| 52 |
model_name,
|
| 53 |
-
torch_dtype=torch.
|
|
|
|
|
|
|
| 54 |
)
|
| 55 |
|
| 56 |
-
|
| 57 |
-
|
|
|
|
| 58 |
|
| 59 |
model_cache[model_name] = {
|
| 60 |
"pipeline": pipe,
|
|
@@ -64,32 +107,40 @@ def load_image_model(model_name):
|
|
| 64 |
return model_cache[model_name]
|
| 65 |
|
| 66 |
def generate_text(prompt, model_name, max_length=100):
|
| 67 |
-
"""Generar texto con el modelo seleccionado"""
|
| 68 |
try:
|
| 69 |
model_data = load_text_model(model_name)
|
| 70 |
tokenizer = model_data["tokenizer"]
|
| 71 |
model = model_data["model"]
|
| 72 |
|
| 73 |
-
#
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
)
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
|
| 94 |
return response
|
| 95 |
|
|
@@ -97,16 +148,22 @@ def generate_text(prompt, model_name, max_length=100):
|
|
| 97 |
return f"Error generando texto: {str(e)}"
|
| 98 |
|
| 99 |
def generate_image(prompt, model_name, num_inference_steps=20):
|
| 100 |
-
"""Generar imagen con el modelo seleccionado"""
|
| 101 |
try:
|
| 102 |
model_data = load_image_model(model_name)
|
| 103 |
pipeline = model_data["pipeline"]
|
| 104 |
|
| 105 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
image = pipeline(
|
| 107 |
prompt,
|
| 108 |
num_inference_steps=num_inference_steps,
|
| 109 |
-
guidance_scale=7.5
|
|
|
|
|
|
|
| 110 |
).images[0]
|
| 111 |
|
| 112 |
return image
|
|
@@ -207,7 +264,7 @@ with gr.Blocks(title="Modelos Libres de IA", theme=gr.themes.Soft()) as demo:
|
|
| 207 |
with gr.Row():
|
| 208 |
with gr.Column():
|
| 209 |
chat_model = gr.Dropdown(
|
| 210 |
-
choices=["
|
| 211 |
value="microsoft/DialoGPT-medium",
|
| 212 |
label="Modelo de Chat"
|
| 213 |
)
|
|
@@ -237,6 +294,35 @@ with gr.Blocks(title="Modelos Libres de IA", theme=gr.themes.Soft()) as demo:
|
|
| 237 |
outputs=[chatbot]
|
| 238 |
)
|
| 239 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 240 |
# Tab de Generación de Imágenes
|
| 241 |
with gr.TabItem("🎨 Generación de Imágenes"):
|
| 242 |
with gr.Row():
|
|
@@ -254,7 +340,7 @@ with gr.Blocks(title="Modelos Libres de IA", theme=gr.themes.Soft()) as demo:
|
|
| 254 |
steps = gr.Slider(
|
| 255 |
minimum=10,
|
| 256 |
maximum=50,
|
| 257 |
-
value=
|
| 258 |
step=5,
|
| 259 |
label="Pasos de inferencia"
|
| 260 |
)
|
|
|
|
| 11 |
MODELS = {
|
| 12 |
"text": {
|
| 13 |
"microsoft/DialoGPT-medium": "Chat conversacional",
|
| 14 |
+
"microsoft/DialoGPT-large": "Chat conversacional avanzado",
|
| 15 |
+
"microsoft/DialoGPT-small": "Chat conversacional rápido",
|
| 16 |
"gpt2": "Generación de texto",
|
| 17 |
+
"gpt2-medium": "GPT-2 mediano",
|
| 18 |
+
"gpt2-large": "GPT-2 grande",
|
| 19 |
"distilgpt2": "GPT-2 optimizado",
|
| 20 |
+
"EleutherAI/gpt-neo-125M": "GPT-Neo pequeño",
|
| 21 |
+
"EleutherAI/gpt-neo-1.3B": "GPT-Neo mediano",
|
| 22 |
+
"microsoft/DialoGPT-medium": "Chat conversacional",
|
| 23 |
+
"facebook/opt-125m": "OPT pequeño",
|
| 24 |
+
"facebook/opt-350m": "OPT mediano",
|
| 25 |
+
"bigscience/bloom-560m": "BLOOM multilingüe",
|
| 26 |
+
"bigscience/bloom-1b1": "BLOOM grande",
|
| 27 |
+
"microsoft/DialoGPT-medium": "Chat conversacional",
|
| 28 |
+
"Helsinki-NLP/opus-mt-es-en": "Traductor español-inglés",
|
| 29 |
+
"Helsinki-NLP/opus-mt-en-es": "Traductor inglés-español"
|
| 30 |
},
|
| 31 |
"image": {
|
| 32 |
"runwayml/stable-diffusion-v1-5": "Stable Diffusion v1.5",
|
| 33 |
+
"CompVis/stable-diffusion-v1-4": "Stable Diffusion v1.4",
|
| 34 |
+
"stabilityai/stable-diffusion-2-1": "Stable Diffusion 2.1",
|
| 35 |
+
"stabilityai/stable-diffusion-xl-base-1.0": "SDXL Base",
|
| 36 |
+
"stabilityai/stable-diffusion-xl-refiner-1.0": "SDXL Refiner",
|
| 37 |
+
"prompthero/openjourney": "Midjourney style",
|
| 38 |
+
"dreamlike-art/dreamlike-photoreal-2.0": "Fotorealista",
|
| 39 |
+
"nitrosocke/Ghibli-Diffusion": "Estilo Studio Ghibli",
|
| 40 |
+
"nitrosocke/mo-di-diffusion": "Estilo moderno",
|
| 41 |
+
"CompVis/stable-diffusion-v1-4": "Stable Diffusion v1.4",
|
| 42 |
+
"runwayml/stable-diffusion-v1-5": "Stable Diffusion v1.5"
|
| 43 |
+
},
|
| 44 |
+
"chat": {
|
| 45 |
+
"microsoft/DialoGPT-medium": "Chat conversacional",
|
| 46 |
+
"microsoft/DialoGPT-large": "Chat conversacional avanzado",
|
| 47 |
+
"microsoft/DialoGPT-small": "Chat conversacional rápido",
|
| 48 |
+
"facebook/opt-350m": "OPT conversacional",
|
| 49 |
+
"bigscience/bloom-560m": "BLOOM multilingüe"
|
| 50 |
}
|
| 51 |
}
|
| 52 |
|
|
|
|
| 54 |
model_cache = {}
|
| 55 |
|
| 56 |
def load_text_model(model_name):
|
| 57 |
+
"""Cargar modelo de texto con soporte para diferentes tipos"""
|
| 58 |
if model_name not in model_cache:
|
| 59 |
print(f"Cargando modelo de texto: {model_name}")
|
|
|
|
|
|
|
| 60 |
|
| 61 |
+
# Detectar tipo de modelo
|
| 62 |
+
if "opus-mt" in model_name.lower():
|
| 63 |
+
# Modelo de traducción
|
| 64 |
+
from transformers import MarianMTModel, MarianTokenizer
|
| 65 |
+
tokenizer = MarianTokenizer.from_pretrained(model_name)
|
| 66 |
+
model = MarianMTModel.from_pretrained(model_name)
|
| 67 |
+
else:
|
| 68 |
+
# Modelo de generación de texto
|
| 69 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 70 |
+
model = AutoModelForCausalLM.from_pretrained(model_name)
|
| 71 |
+
|
| 72 |
+
# Configurar para chat si es DialoGPT
|
| 73 |
+
if "dialogpt" in model_name.lower():
|
| 74 |
+
tokenizer.pad_token = tokenizer.eos_token
|
| 75 |
+
model.config.pad_token_id = model.config.eos_token_id
|
| 76 |
|
| 77 |
model_cache[model_name] = {
|
| 78 |
"tokenizer": tokenizer,
|
|
|
|
| 83 |
return model_cache[model_name]
|
| 84 |
|
| 85 |
def load_image_model(model_name):
|
| 86 |
+
"""Cargar modelo de imagen - optimizado para velocidad"""
|
| 87 |
if model_name not in model_cache:
|
| 88 |
print(f"Cargando modelo de imagen: {model_name}")
|
| 89 |
+
|
| 90 |
+
# Optimizaciones para CPU y velocidad
|
| 91 |
pipe = StableDiffusionPipeline.from_pretrained(
|
| 92 |
model_name,
|
| 93 |
+
torch_dtype=torch.float32, # Usar float32 para CPU
|
| 94 |
+
safety_checker=None, # Desactivar safety checker para velocidad
|
| 95 |
+
requires_safety_checker=False
|
| 96 |
)
|
| 97 |
|
| 98 |
+
# Optimizaciones adicionales
|
| 99 |
+
pipe.enable_attention_slicing() # Reducir uso de memoria
|
| 100 |
+
pipe.enable_sequential_cpu_offload() # Optimizar para CPU
|
| 101 |
|
| 102 |
model_cache[model_name] = {
|
| 103 |
"pipeline": pipe,
|
|
|
|
| 107 |
return model_cache[model_name]
|
| 108 |
|
| 109 |
def generate_text(prompt, model_name, max_length=100):
|
| 110 |
+
"""Generar texto con el modelo seleccionado - mejorado para diferentes tipos"""
|
| 111 |
try:
|
| 112 |
model_data = load_text_model(model_name)
|
| 113 |
tokenizer = model_data["tokenizer"]
|
| 114 |
model = model_data["model"]
|
| 115 |
|
| 116 |
+
# Detectar si es modelo de traducción
|
| 117 |
+
if "opus-mt" in model_name.lower():
|
| 118 |
+
# Traducción
|
| 119 |
+
inputs = tokenizer.encode(prompt, return_tensors="pt", max_length=512, truncation=True)
|
| 120 |
+
with torch.no_grad():
|
| 121 |
+
outputs = model.generate(inputs, max_length=max_length, num_beams=4, early_stopping=True)
|
| 122 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 123 |
+
else:
|
| 124 |
+
# Generación de texto
|
| 125 |
+
inputs = tokenizer.encode(prompt, return_tensors="pt")
|
| 126 |
+
|
| 127 |
+
# Generar
|
| 128 |
+
with torch.no_grad():
|
| 129 |
+
outputs = model.generate(
|
| 130 |
+
inputs,
|
| 131 |
+
max_length=max_length,
|
| 132 |
+
num_return_sequences=1,
|
| 133 |
+
temperature=0.7,
|
| 134 |
+
do_sample=True,
|
| 135 |
+
pad_token_id=tokenizer.eos_token_id
|
| 136 |
+
)
|
| 137 |
+
|
| 138 |
+
# Decodificar respuesta
|
| 139 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 140 |
+
|
| 141 |
+
# Para DialoGPT, extraer solo la respuesta del asistente
|
| 142 |
+
if "dialogpt" in model_name.lower():
|
| 143 |
+
response = response.replace(prompt, "").strip()
|
| 144 |
|
| 145 |
return response
|
| 146 |
|
|
|
|
| 148 |
return f"Error generando texto: {str(e)}"
|
| 149 |
|
| 150 |
def generate_image(prompt, model_name, num_inference_steps=20):
|
| 151 |
+
"""Generar imagen con el modelo seleccionado - optimizado para velocidad"""
|
| 152 |
try:
|
| 153 |
model_data = load_image_model(model_name)
|
| 154 |
pipeline = model_data["pipeline"]
|
| 155 |
|
| 156 |
+
# Optimizaciones para velocidad
|
| 157 |
+
if num_inference_steps > 20:
|
| 158 |
+
num_inference_steps = 20 # Limitar a máximo 20 pasos para velocidad
|
| 159 |
+
|
| 160 |
+
# Generar imagen con configuración optimizada
|
| 161 |
image = pipeline(
|
| 162 |
prompt,
|
| 163 |
num_inference_steps=num_inference_steps,
|
| 164 |
+
guidance_scale=7.0, # Reducido de 7.5 para velocidad
|
| 165 |
+
height=512, # Tamaño fijo para consistencia
|
| 166 |
+
width=512
|
| 167 |
).images[0]
|
| 168 |
|
| 169 |
return image
|
|
|
|
| 264 |
with gr.Row():
|
| 265 |
with gr.Column():
|
| 266 |
chat_model = gr.Dropdown(
|
| 267 |
+
choices=list(MODELS["chat"].keys()),
|
| 268 |
value="microsoft/DialoGPT-medium",
|
| 269 |
label="Modelo de Chat"
|
| 270 |
)
|
|
|
|
| 294 |
outputs=[chatbot]
|
| 295 |
)
|
| 296 |
|
| 297 |
+
# Tab de Traducción
|
| 298 |
+
with gr.TabItem("🌐 Traducción"):
|
| 299 |
+
with gr.Row():
|
| 300 |
+
with gr.Column():
|
| 301 |
+
translate_model = gr.Dropdown(
|
| 302 |
+
choices=["Helsinki-NLP/opus-mt-es-en", "Helsinki-NLP/opus-mt-en-es"],
|
| 303 |
+
value="Helsinki-NLP/opus-mt-es-en",
|
| 304 |
+
label="Modelo de Traducción"
|
| 305 |
+
)
|
| 306 |
+
translate_text = gr.Textbox(
|
| 307 |
+
label="Texto a traducir",
|
| 308 |
+
placeholder="Escribe el texto que quieres traducir...",
|
| 309 |
+
lines=3
|
| 310 |
+
)
|
| 311 |
+
translate_btn = gr.Button("Traducir", variant="primary")
|
| 312 |
+
|
| 313 |
+
with gr.Column():
|
| 314 |
+
translate_output = gr.Textbox(
|
| 315 |
+
label="Traducción",
|
| 316 |
+
lines=3,
|
| 317 |
+
interactive=False
|
| 318 |
+
)
|
| 319 |
+
|
| 320 |
+
translate_btn.click(
|
| 321 |
+
generate_text,
|
| 322 |
+
inputs=[translate_text, translate_model, gr.Slider(value=100, visible=False)],
|
| 323 |
+
outputs=translate_output
|
| 324 |
+
)
|
| 325 |
+
|
| 326 |
# Tab de Generación de Imágenes
|
| 327 |
with gr.TabItem("🎨 Generación de Imágenes"):
|
| 328 |
with gr.Row():
|
|
|
|
| 340 |
steps = gr.Slider(
|
| 341 |
minimum=10,
|
| 342 |
maximum=50,
|
| 343 |
+
value=15,
|
| 344 |
step=5,
|
| 345 |
label="Pasos de inferencia"
|
| 346 |
)
|