Is the DeepONet predict solution more efficient than the original method?

[xuliang5115]

I generage a dataset from original method( finite difference methods for 2-D wave problem)，train a DeepONet and predict solutions.
But, it seems not to speed the predict efficient compare to original method.
I notice many article about DeepONet has the 1000x speed up than original method.
What can i do to improve it?

Thanks a lot for your help.