Update app.py

app.py

@@ -36,7 +36,7 @@ if not path.exists(cache_path):
 
 pipe = StableDiffusionXLPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.bfloat16)
 pipe.to(device="cpu", dtype=torch.bfloat16)
-unet_state = load_file(hf_hub_download("ByteDance/Hyper-SD", "Hyper-SDXL-1step-Unet.safetensors"), device="cuda")
+unet_state = load_file(hf_hub_download("ByteDance/Hyper-SD", "Hyper-SDXL-1step-Unet.safetensors"), device="cpu")
 pipe.unet.load_state_dict(unet_state)
 pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config, timestep_spacing ="trailing")
 
@@ -59,7 +59,7 @@ with gr.Blocks() as demo:
             @spaces.GPU
             def process_image(num_images, height, width, steps, guidance, prompt, seed):
                 global pipe
-                with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16), timer("inference"):
+                with torch.inference_mode(), torch.autocast("cpu", dtype=torch.bfloat16), timer("inference"):
                     return pipe(
                         prompt=[prompt]*num_images,
                         generator=torch.Generator().manual_seed(int(seed)),