Great, I am starting to get hang of it. Sorry for using this thread documenting my learning, but in the documentation the summary of the algorithm finally let me understand the parameters. So if I want to have more samples from the posterior I will indeed increase draws parameter but for debugging purposes if I want to get the annealing process done faster I should reduce the threshold parameter. SMC attempts to get an effective sample size of \mathrm{N}_\mathrm{eff}=draws\timesthreshold at each stage of the annealing.
Will I run into problems with SMC if I write the simulator using numba and gpu acceleration?
Thank you for your help 