Open birsenc opened 2 weeks ago
@birsenc thank you for reaching out. In your case GPUs won't help. They are just useful for training and model building. What you need is to parallemize it over the number of stations (and perhaps months) using many more CPUs. I have implemented a crude parallel process (this is what the number of CPU parameter is for) but you may improve it by running a higher order parallelization.
Dear @smousavi05 and all,
We are running EQTransformer at HPC using CPU. We have more than 100 seismic stations and 3 years of data. We want to accelerate the speed of predicition. I have questions related to this issue.
Thank you very much in advance
Do we have to choose the below parameters and use exactly the same values at the slurm also?
gpuid: int, default=None Id of GPU used for the prediction. If using CPU set to None.
gpu_limit: int, default=None Set the maximum percentage of memory usage for the GPU.
number_of_cpus: int, default=5 Number of CPUs used for the parallel preprocessing and feeding of data for prediction.
use_multiprocessing: bool, default=True If True, multiple CPUs will be used for the preprocessing of data even when GPU is used for the prediction.