I am trying to fix a complex GAM with temporal autocorrelation in brms (version 2.9.0, running on Ubuntu). As the model is fairly complex, I noticed that it would be better to split the chains on multiple cores. This would enable to run let’s say 4 MCMC chains on a workstation with a processor of about 40 cores.
I noticed that in brms the number of cores is automatically set up to equate the number of chains. Is there a way to split 3-4 chains on a high number of cores? In STAN there is the “map_rect” command to do this: is there something similar on brms?