I’m running Stan on an HPC cluster and I can allocate up to 128 GB per core. When I run Stan on this core, I am only running one chain at a time and it only uses about 7 GB of RAM. Is there a way to tell Stan to use more memory? Or is this as much as it will use?
I’m using rstan, but could switch to a different wrapper if it would be more effective.
Stan will only use as much memory as required to run the sampler, and in the case of RStan (or PyStan) store the draws.
You can use memory.limit() to report the running memory, and use memory.size(…) function to maximum memory allocation