load in advance

app.py

@@ -15,11 +15,12 @@ model_ckpt = "stabilityai/stable-diffusion-xl-base-1.0"
 model_ckpt_turbo = "stabilityai/sdxl-turbo"
 pipe = StableDiffusionXLPipeline.from_pretrained(model_ckpt, torch_dtype=dtype).to(device)
 pipe_turbo = StableDiffusionXLPipeline_Turbo.from_pretrained(model_ckpt_turbo, torch_dtype=dtype).to(device)
-generator = torch.Generator(device=dtype)
+generator = torch.Generator(device='cuda')
 torch.cuda.empty_cache()
 
 @spaces.GPU(duration=100)
 def infer_gpu_part(pipe, seed, prompt, negative_prompt, ddim_steps, guidance_scale, resolutions_list, fast_mode, cosine_scale, disable_freeu, restart_steps):
+    generator = torch.Generator(device='cuda')
     generator = generator.manual_seed(seed)
     if not disable_freeu:
         register_free_upblock2d(pipe, b1=1.1, b2=1.2, s1=0.6, s2=0.4)
@@ -33,6 +34,7 @@ def infer_gpu_part(pipe, seed, prompt, negative_prompt, ddim_steps, guidance_sca
 
 @spaces.GPU(duration=20)
 def infer_gpu_part_turbo(pipe, seed, prompt, negative_prompt, ddim_steps, guidance_scale, resolutions_list, fast_mode, cosine_scale, disable_freeu, restart_steps):
+    generator = torch.Generator(device='cuda')
     generator = generator.manual_seed(seed)
     if not disable_freeu:
         register_free_upblock2d(pipe, b1=1.1, b2=1.2, s1=0.6, s2=0.4)