How to use pre-allocated buffer in output ?

[Jsy0220]

Hi, I tried to use pre-allocated buffer by set_output_tensor of InferRequest, but it seems not allowed Can't SetBlob with name: Clamp_1431, because model output and blob are incompatible..
The IR model I used has dynamic inputs and output. I used tool mo to convert it from onnx and load it in c++, try to set pre-allocated buffer for output tensor when inferencing but failed.
So what's the problem ?
Thanks