Closed gbrunin closed 1 year ago
Hopefully closed by setting lots of TF_*
env vars so that TF does not hog all of a CPU at once
Yep, seems to work. Thanks!
For anyone having a similar issue, try setting the environment variables:
export OPENBLAS_NUM_THREADS=1
export MKL_NUM_THREADS=1
export OMP_NUM_THREADS=1
export TF_NUM_INTRAOP_THREADS=1
export TF_NUM_INTEROP_THREADS=1
export CUDA_VISIBLE_DEVICES=1
export CUDA_VISIBLE_DEVICES=1
I think this should be -1
? Probably this is equivalent unless you have more than 1 GPU on your system...
Yes ! No need of a cuda device
When using FitGenetic, I sometimes have the following issue: the process stops around the end of an iteration, without throwing an error, and without stopping to run. It's not reproducible, but happens quite often in my case. I run the code on my laptop, in a
main()
method. When it happens, I simply re-run the code and it works (or not, depends on the phase of the moon). Pierre-Paul suspects a problem with parallelization, more precisely a child process that never returns its result. Any clue @ml-evs ?