Update app.py

app.py

@@ -18,7 +18,9 @@ conda_bin = os.path.join(conda_prefix, "bin")
 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-nvcc')
 
 import random
 import uuid
@@ -185,7 +187,7 @@ def load_and_prepare_model():
     return pipe
     
 hidet.option.parallel_build(True)
-hidet.option.parallel_tune(1,2.0)
+#hidet.option.parallel_tune(1,2.0)
 #hidet.torch.dynamo_config.steal_weights(False)
 torch._dynamo.config.suppress_errors = True
 torch._dynamo.disallow_in_graph(diffusers.models.attention.BasicTransformerBlock)
@@ -199,7 +201,7 @@ hidet.option.cache_dir("local_cache")
 # use float16 data type as the accumulate data type in operators with reduction
 #hidet.torch.dynamo_config.use_fp16_reduction(True)
 # use tensorcore
-hidet.torch.dynamo_config.use_tensor_core()
+#hidet.torch.dynamo_config.use_tensor_core()
 hidet.torch.dynamo_config.steal_weights(False)
 
 # Preload and compile both models