Hi,
I am using the debug_executor of TVM to get detailed latency information of my network. tvm_debug_module = debug_executor.create(graph, tvm_remote_model, tvm_remote_device, dump_root=’./logs/autolog/tvm-debugger/’) profile_data = tvm_debug_module.profile() profiling_json = profile_data.json() profile_dict = json.loads(profiling_json) calls = profile_dict[“calls”]
I wanted to map these latency information to the original layers of the network using the layer shapes. How can I interpret the Argument Shapes that are included in the calls variable? I saw that they are different depending on the type of the node but couldn’t find any source explaining these shapes.