Model size are not reduced after simplification

[junde-cadence]

I tried to simplify TinyLlama with the code, but the simplified onnx file is almost with the same size with non-simplified one. It is appreciated if you can provide onnx sizes of the original Llama onnx model and the one after simplification.

[luchangli03]

maybe you export onnx model without using const folding

[luchangli03]

please try the new code