Update app.py

app.py

@@ -20,6 +20,8 @@ os.environ["PATH"] = conda_bin + os.pathsep + os.environ["PATH"]
 # Activate the base environment (adjust if needed)
 os.system(f'{conda_bin}/conda init  --all') 
 os.system(f'{conda_bin}/conda activate base') 
+os.system(f'{conda_bin}/conda install nvidia/label/cuda-12.4.0::cuda-nvrtc')
+os.system(f'{conda_bin}/conda install nvidia/label/cuda-12.4.0::cuda-nvrtc-dev')
 os.system(f'{conda_bin}/conda install nvidia/label/cuda-12.4.0::cuda-cudart')
 os.system(f'{conda_bin}/conda install nvidia/label/cuda-12.4.0::cuda-cudart-dev')
 os.system(f'{conda_bin}/conda install nvidia/label/cuda-12.4.0::cuda-nvcc')
@@ -190,7 +192,7 @@ def load_and_prepare_model():
     return pipe
     
 hidet.option.parallel_build(True)
-hidet.option.parallel_tune(-1,1.75)
+hidet.option.parallel_tune(2,3.0)
 #hidet.torch.dynamo_config.steal_weights(False)
 torch._dynamo.config.suppress_errors = True
 torch._dynamo.disallow_in_graph(diffusers.models.attention.BasicTransformerBlock)