Concern about the running time

[xhsoldier]

The output model is about 20MB, which is too big for the mobile phone to running. The next point prediction should be executed in a very short of time.

Do you have any idea on this? Compress the model may result worse accuracy.