Update app.py

app.py

@@ -20,10 +20,16 @@ available_models = [
 MAX_SEED = np.iinfo(np.int32).max
 MAX_IMAGE_SIZE = 1024
 
+
 def init_model(model_repo_id):
     torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
     pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-    return pipe.to(device)
+    return pipe
+    
+loaded_models = {}
+
+for model in available_models:
+    loaded_models[model] = init_model(model)
 
 # @spaces.GPU #[uncomment to use ZeroGPU]
 def infer(
@@ -38,7 +44,7 @@ def infer(
     num_inference_steps,
     progress=gr.Progress(track_tqdm=True),
 ):
-    pipe = init_model(model_repo_id)
+    pipe = loaded_models[model_repo_id].to(device)
     
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)