Update app.py

app.py

@@ -7,7 +7,6 @@ from diffusers.image_processor import VaeImageProcessor
 from transformers import CLIPImageProcessor
 from huggingface_hub import hf_hub_download
 from safetensors.torch import load_file
-from PIL import Image
 
 device = "cuda"
 dtype = torch.float16
@@ -21,30 +20,24 @@ opts = {
     "8 Steps" : ("sdxl_lightning_8step_unet.safetensors", 8),
 }
 
-# Default to load 4-step model.
-step_loaded = 4
-unet = UNet2DConditionModel.from_config(base, subfolder="unet")
-unet.load_state_dict(load_file(hf_hub_download(repo, opts["4 Steps"][0])))
-pipe = StableDiffusionXLPipeline.from_pretrained(base, unet=unet, torch_dtype=dtype, variant="fp16")
-pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
-
-# Safety checker.
-safety_checker=StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker")
-feature_extractor=CLIPImageProcessor.from_pretrained("openai/clip-vit-base-patch32")
-image_processor = VaeImageProcessor(vae_scale_factor=8)
-
 # Inference function.
-@spaces.GPU(enable_queue=True)
+@spaces.GPU()
 def generate(prompt, option, progress=gr.Progress()):
-    global step_loaded
     print(prompt, option)
     ckpt, step = opts[option]
+
+    # Main pipeline.
+    unet = UNet2DConditionModel.from_config(base, subfolder="unet")
+    unet.load_state_dict(load_file(hf_hub_download(repo, opts["4 Steps"][0]))).to(device, dtype)
+    pipe = StableDiffusionXLPipeline.from_pretrained(base, unet=unet, torch_dtype=dtype, variant="fp16")
+    pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing").to(device, dtype)
+
+    # Safety checker.
+    safety_checker=StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker").to(device, dtype)
+    feature_extractor=CLIPImageProcessor.from_pretrained("openai/clip-vit-base-patch32")
+    image_processor = VaeImageProcessor(vae_scale_factor=8)
+
     progress((0, step))
-    
-    if pipe.device.type != device:
-        pipe.to(device, dtype)
-    if safety_checker.device.type != device:
-        safety_checker.to(device, dtype)
 
     if step != step_loaded:
         print(f"Switching checkpoint from {step_loaded} to {step}")