Closed arendu-zz closed 5 years ago
Thanks Adi. What's the effect of using -q gpu.q
vs not using on the CLSP grid (with the -l gpu=1
in effect)?
adding the -q gpu.q
makes the job stay in wait state indefinitely... I think because the CLSP grid does not recognize gpu.q
maybe worth pointing to a newer gpu (it crashed for me when the grid allocated b04 with an old gpu)
qsub -S /bin/bash -V -cwd -l gpu=1,hostname=b1[12345678]*|c*,h_rt=24:00:00,num_proc=2 -j y ../../scripts/train.sh -p tiny_rnn_gpu.hpm -e sockeye_gpu
ok. merged.
added a line for running quickstart on gpu on CLSP grid. (i.e. removed the
-g gpu.p
flag)