it doesn't include weight input information in the profiling trace

[andyzheng-snps]

I want to get the input and output shape information from profiling trace. however when I check the profiling trace and found one of input doesn't appear in this trace. for example, below MatMul, only one input appears in this log. I need get another input from model initializer. This brings more effort to identify which input missing in this trace log.

"output_type_shape" : [{"float":[8,64,50257]}],"output_size" : "102926336","parameter_size" : "0","activation_size" : "1572864","node_index" : "1755","input_type_shape" : [{"float":[8,64,768]}],"provider" : "CPUExecutionProvider","op_name" : "MatMul"}},

From function: CalculateTotalInputSizes in the below link, it will loop all of inputs to determine if the input is parameter or activation. it should cover the weight input. That is my confusion.
https://github.com/microsoft/onnxruntime/blob/main/onnxruntime/core/framework/sequential_executor.cc

  if (is_param) {
    input_parameter_sizes += tensor_size;
  } else {
    input_activation_sizes += tensor_size;
  }

Anybody can help me clarify this question?

and another question, in the training mode, it can still generate this kind of profiling trace?

Andy