Commit ·
5e13943
1
Parent(s): 4ab181b
first commit
Browse files- handler.py +49 -0
- requirements.txt +7 -0
handler.py
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from typing import List, Any
|
| 2 |
+
import torch
|
| 3 |
+
from diffusers import StableCascadePriorPipeline, StableCascadeDecoderPipeline
|
| 4 |
+
|
| 5 |
+
# Configurar el dispositivo para ejecutar el modelo
|
| 6 |
+
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
| 7 |
+
if device.type != 'cuda':
|
| 8 |
+
raise ValueError("Se requiere ejecutar en GPU")
|
| 9 |
+
|
| 10 |
+
# Configurar el tipo de dato mixto basado en la capacidad de la GPU
|
| 11 |
+
dtype = torch.bfloat16 if torch.cuda.get_device_capability(device.index)[0] >= 8 else torch.float16
|
| 12 |
+
|
| 13 |
+
class EndpointHandler():
|
| 14 |
+
def __init__(self):
|
| 15 |
+
# Inicializar aquí si es necesario
|
| 16 |
+
pass
|
| 17 |
+
|
| 18 |
+
def __call__(self, data: Any) -> List[Any]:
|
| 19 |
+
# Configurar el número de imágenes por prompt
|
| 20 |
+
num_images_per_prompt = 1
|
| 21 |
+
|
| 22 |
+
# Cargar los modelos con el tipo de dato y dispositivo correctos
|
| 23 |
+
prior = StableCascadePriorPipeline.from_pretrained("stabilityai/stable-cascade-prior", torch_dtype=dtype).to(device)
|
| 24 |
+
decoder = StableCascadeDecoderPipeline.from_pretrained("stabilityai/stable-cascade", torch_dtype=dtype).to(device)
|
| 25 |
+
|
| 26 |
+
prompt = data.get("inputs", "Una imagen interesante") # Asegúrate de pasar un prompt adecuado
|
| 27 |
+
negative_prompt = data.get("negative_prompt", "")
|
| 28 |
+
|
| 29 |
+
prior_output = prior(
|
| 30 |
+
prompt=prompt,
|
| 31 |
+
height=512,
|
| 32 |
+
width=512,
|
| 33 |
+
negative_prompt=negative_prompt,
|
| 34 |
+
guidance_scale=7.5,
|
| 35 |
+
num_inference_steps=50,
|
| 36 |
+
num_images_per_prompt=num_images_per_prompt,
|
| 37 |
+
)
|
| 38 |
+
|
| 39 |
+
decoder_output = decoder(
|
| 40 |
+
image_embeddings=prior_output["image_embeddings"].half(),
|
| 41 |
+
prompt=prompt,
|
| 42 |
+
negative_prompt=negative_prompt,
|
| 43 |
+
guidance_scale=7.5,
|
| 44 |
+
output_type="pil",
|
| 45 |
+
num_inference_steps=20
|
| 46 |
+
)
|
| 47 |
+
|
| 48 |
+
# Asumiendo que quieres retornar la primera imagen
|
| 49 |
+
return [decoder_output.images[0]]
|
requirements.txt
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
safetensors
|
| 2 |
+
opencv-python
|
| 3 |
+
controlnet_hinter==0.0.5
|
| 4 |
+
git+https://github.com/kashif/diffusers.git@diffusers-yield-callback
|
| 5 |
+
https://gradio-builds.s3.amazonaws.com/aabb08191a7d94d2a1e9ff87b0d3c3987cd519c5/gradio-4.18.0-py3-none-any.whl
|
| 6 |
+
accelerate
|
| 7 |
+
transformers
|