Update app.py

app.py

@@ -32,7 +32,7 @@ os.system(f'{conda_bin}/conda install nvidia/label/cuda-12.4.0::cuda-nvcc')
 os.system(f'{conda_bin}/conda install pytorch::pytorch-cuda')
 #os.system(f'{conda_bin}/conda install rcdr_py37::tensorrt')
 #subprocess.run(['sh', './hidet.sh'])
-#import hidet
+import hidet
 #print(dir(hidet)) 
 
 import random
@@ -41,7 +41,6 @@ import gradio as gr
 import numpy as np
 from PIL import Image
 
-
 import diffusers
 from diffusers import AutoencoderKL, StableDiffusionXLPipeline
 from diffusers import EulerAncestralDiscreteScheduler
@@ -195,12 +194,12 @@ def load_and_prepare_model():
     
     #pipe.unet.to(memory_format=torch.channels_last)
     #pipe.enable_vae_tiling()
-    #pipe.unet = torch.compile(pipe.unet, backend="hidet")
-    pipe.unet = torch.compile(pipe.unet, backend="cudagraphs", dynamic=False)
+    pipe.unet = torch.compile(pipe.unet, backend="hidet")
+    #pipe.unet = torch.compile(pipe.unet, backend="cudagraphs", dynamic=False)
     pipe.to(device=device, dtype=torch.bfloat16)
 
     return pipe
-'''
+
 hidet.option.parallel_build(True)
 hidet.option.parallel_tune(2,2.0)
 torch._dynamo.config.suppress_errors = True
@@ -217,7 +216,7 @@ hidet.option.cache_dir("local_cache")
 # use tensorcore
 hidet.torch.dynamo_config.use_tensor_core()
 hidet.torch.dynamo_config.steal_weights(False)
-'''
+
 # Preload and compile both models
 
 pipe = load_and_prepare_model()