Update app.py

app.py

@@ -99,25 +99,25 @@ def load_and_prepare_model():
        # low_cpu_mem_usage = False,
         add_watermarker=False,
     )
-    pipe.vae = vaeXL.to(torch.bfloat16)
+    pipe.vae = vaeXL #.to(torch.bfloat16)
     pipe.scheduler = sched
     #pipe.vae.do_resize=False
     #pipe.vae.vae_scale_factor=8
-    #pipe.to(device=device, dtype=torch.bfloat16)
-    pipe.to(device)
-    pipe.to(torch.bfloat16)
+    #pipe.to(device)
+    #pipe.to(torch.bfloat16)
     pipe.vae.set_default_attn_processor()
     print(f'init noise scale: {pipe.scheduler.init_noise_sigma}')
     pipe.watermark=None
     pipe.safety_checker=None
     pipe.unet = pipe.unet.to(memory_format=torch.contiguous_format)
     pipe.unet = torch.compile(pipe.unet, backend="hidet")
+    pipe.to(device=device, dtype=torch.bfloat16)
 
     return pipe
     
           # for compile
 hidet.option.parallel_build(True)
-hidet.option.parallel_tune(-1,4.0)
+hidet.option.parallel_tune(-1,8.0)
 torch._dynamo.config.suppress_errors = True
 torch._dynamo.disallow_in_graph(diffusers.models.attention.BasicTransformerBlock)
 # more search