parallel SGP training

[TSMtang]

Dear Flare developers,

I've been using FLARE for training force field. I noticed that when using Sparse Gaussian Process regression (gp: SGP_Wrapper), there is no option to setup the number of CPUs (e.g. n_cpus: 10 is not working). I wonder if there is any way I can use SGP_Wrapper with multiple processors?

Thank you in advance
Shengming Tang

[YuuuXie]

Hi @TSMtang , in SGP_Wrapper you don't need to specify number of CPUs. Instead you can set the environmental variable:

export OMP_NUM_THREADS=<number_of_cpus>

before you run your training. The number of CPUs has to be equal or smaller than the number on a single node, since we use OpenMP to parallelize the SGP