Update app.py

app.py

@@ -62,7 +62,7 @@ checkpoint = "microsoft/Phi-3.5-mini-instruct"
 #vae = AutoencoderKL.from_pretrained("ford442/sdxl-vae-bf16")
 vaeXL = AutoencoderKL.from_pretrained("stabilityai/sdxl-vae", safety_checker=None, use_safetensors=False) #, device_map='cpu') #.to(torch.bfloat16) #.to(device=device, dtype=torch.bfloat16)
 
-pipe = StableDiffusion3Pipeline.from_pretrained("ford442/stable-diffusion-3.5-medium-bf16") #.to(torch.bfloat16)
+pipe = StableDiffusion3Pipeline.from_pretrained("ford442/stable-diffusion-3.5-medium-bf16").to(device=device, dtype=torch.bfloat16)
 #pipe = StableDiffusion3Pipeline.from_pretrained("ford442/stable-diffusion-3.5-medium-bf16").to(torch.device("cuda:0"))
 #pipe = StableDiffusion3Pipeline.from_pretrained("ford442/RealVis_Medium_1.0b_bf16", torch_dtype=torch.bfloat16)
 #pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3.5-medium", token=hftoken, torch_dtype=torch.float32, device_map='balanced')
@@ -81,6 +81,8 @@ refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained("ford442/stable-diffu
 refiner.scheduler=EulerAncestralDiscreteScheduler.from_config(refiner.scheduler.config)
 #refiner.enable_model_cpu_offload()
 
+#pipe.to(device=device, dtype=torch.bfloat16)
+
 #refiner.scheduler.config.requires_aesthetics_score=False
 #refiner.to(device)
 #refiner = torch.compile(refiner)
@@ -133,7 +135,6 @@ def infer(
     progress=gr.Progress(track_tqdm=True),
 ):
     upscaler_2.to(torch.device('cpu'))
-    pipe.to(device=device, dtype=torch.bfloat16)
     torch.set_float32_matmul_precision("highest")
     seed = random.randint(0, MAX_SEED)
     generator = torch.Generator(device='cuda').manual_seed(seed)