Fix a bug with deepcopy by saving and reloading model insted

[bogdanteleaga]

We found out that deepcopy sometimes breaks models in unexpected ways, so this is a workaround fix that should achieve the same result. But I don't think it's ideal, let me know if you have any better ideas @christhetree @hyakuchiki or let's merge it as is.

For example on the DDSP models

(Pdb) sqw.w2w_base.agg_params
tensor([[0.5000],
        [0.5000],
        [0.5000],
        [0.5000]])
(Pdb) script.w2w_base.agg_params
tensor([[0.5000],
        [0.5000],
        [0.5000],
        [0.5000]])
(Pdb) copy.deepcopy(script).w2w_base.agg_params
tensor([0.5000])

[christhetree]

I can't think of a better workaround for now, so fine with me as well.