Update app.py

app.py

@@ -13,22 +13,21 @@ from transformers import (
 )
 from diffusers import VQModel
 import gradio as gr
+import spaces 
 
 device = 'cuda' if torch.cuda.is_available() else 'cpu'
 
-def load_model():
-    model_path = "MeissonFlow/Meissonic"
-    model = Transformer2DModel.from_pretrained(model_path, subfolder="transformer")
-    vq_model = VQModel.from_pretrained(model_path, subfolder="vqvae")
-    text_encoder = CLIPTextModelWithProjection.from_pretrained(model_path, subfolder="text_encoder")
-    tokenizer = CLIPTokenizer.from_pretrained(model_path, subfolder="tokenizer")
-    scheduler = Scheduler.from_pretrained(model_path, subfolder="scheduler")
-    pipe = Pipeline(vq_model, tokenizer=tokenizer, text_encoder=text_encoder, transformer=model, scheduler=scheduler)
-    return pipe.to(device)
-
-pipe = load_model()
-
-def generate_image(prompt, negative_prompt, resolution, steps, cfg):
+model_path = "MeissonFlow/Meissonic"
+model = Transformer2DModel.from_pretrained(model_path, subfolder="transformer")
+vq_model = VQModel.from_pretrained(model_path, subfolder="vqvae")
+text_encoder = CLIPTextModelWithProjection.from_pretrained(model_path, subfolder="text_encoder")
+tokenizer = CLIPTokenizer.from_pretrained(model_path, subfolder="tokenizer")
+scheduler = Scheduler.from_pretrained(model_path, subfolder="scheduler")
+pipe = Pipeline(vq_model, tokenizer=tokenizer, text_encoder=text_encoder, transformer=model, scheduler=scheduler)
+pipe.to(device)
+
+@spaces.GPU
+def generate_image(prompt, negative_prompt, resolution, steps, cfg, progress=gr.Progress(track_tqdm=True)):
     image = pipe(
         prompt=prompt,
         negative_prompt=negative_prompt,