Update app.py

app.py

@@ -202,7 +202,7 @@ def load_and_prepare_model():
     #pipe.unet = torch.compile(pipe.unet, backend="hidet", dynamic=False, mode='max-autotune-no-cudagraphs') #.to(device=device, dtype=torch.bfloat16)
     #pipe.unet = torch.compile(pipe.unet, backend="hidet", dynamic=False, options={'epilogue_fusion': True, 'shape_padding': True}) #.to(device=device, dtype=torch.bfloat16)
     #pipe.unet = torch.compile(pipe.unet, backend="hidet")
-    pipe.unet = torch.compile(pipe.unet, backend="hidet", dynamic=False)
+    pipe.unet = torch.compile(pipe.unet, backend="hidet", dynamic=False, options={"search_space": 0})
     #pipe.unet = torch.compile(pipe.unet, backend="torch_tensorrt", dynamic=False, options={"precision": torch.bfloat16,"optimization_level": 4,})
     pipe.to(device=device, dtype=torch.bfloat16)
 
@@ -210,11 +210,11 @@ def load_and_prepare_model():
 
 hidet.option.parallel_build(True)
 hidet.option.parallel_tune(2,2.0)
-torch._dynamo.config.suppress_errors = True
+#torch._dynamo.config.suppress_errors = True
 torch._dynamo.disallow_in_graph(diffusers.models.attention.BasicTransformerBlock)
 
 # more search
-hidet.torch.dynamo_config.search_space(0)
+#hidet.torch.dynamo_config.search_space(0)
 #hidet.torch.dynamo_config.dump_graph_ir("./local_graph")
 hidet.option.cache_dir("local_cache")
 # automatically transform the model to use float16 data type
@@ -222,8 +222,8 @@ 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.steal_weights(False)
+#hidet.torch.dynamo_config.use_tensor_core()
+#hidet.torch.dynamo_config.steal_weights(False)
 
 # Preload and compile both models