From 3d941d80e4e5fb733fb7ecafaaacab4e6451e7cd Mon Sep 17 00:00:00 2001 From: Gregor von Laszewski Date: Sun, 15 Oct 2023 10:00:25 -0400 Subject: [PATCH] update lsf --- .../cloudmask/target/summit-no-sciml/16.lsf | 2 +- .../cloudmask/target/summit-no-sciml/Makefile | 23 +++++++++++++++++++ .../target/summit-no-sciml/cloud_GPU_1.job | 2 +- .../target/summit-no-sciml/cloud_GPU_12.job | 2 +- .../target/summit-no-sciml/cloud_GPU_18.job | 2 +- .../target/summit-no-sciml/cloud_GPU_2.job | 5 +++- .../target/summit-no-sciml/cloud_GPU_24.job | 2 +- .../target/summit-no-sciml/cloud_GPU_30.job | 2 +- .../target/summit-no-sciml/cloud_GPU_4.job | 2 +- .../target/summit-no-sciml/cloud_GPU_6.job | 2 +- 10 files changed, 35 insertions(+), 9 deletions(-) diff --git a/benchmarks/cloudmask/target/summit-no-sciml/16.lsf b/benchmarks/cloudmask/target/summit-no-sciml/16.lsf index f1fcff15..7f46d586 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/16.lsf +++ b/benchmarks/cloudmask/target/summit-no-sciml/16.lsf @@ -13,7 +13,7 @@ module purge module load open-ce #module load ibm-wml-ce - +export WORDDIR=/gpfs/alpine/gen150/proj-shared/$USER source $WORKDIR/ENV3/bin/activate #This runs on many nodes diff --git a/benchmarks/cloudmask/target/summit-no-sciml/Makefile b/benchmarks/cloudmask/target/summit-no-sciml/Makefile index 48759742..b4f03476 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/Makefile +++ b/benchmarks/cloudmask/target/summit-no-sciml/Makefile @@ -5,5 +5,28 @@ bsub cloud_GPU_2.job # bsub 2.lsf +4: + bsub cloud_GPU_4.job + # bsub 4.lsf +6: + bsub cloud_GPU_6.job + # bsub 6.lsf + +12: + bsub cloud_GPU_12.job + # bsub 12.lsf + +18: + bsub cloud_GPU_18.job + # bsub 18.lsf + +24: + bsub cloud_GPU_24.job + # bsub 24.lsf + +30: + bsub cloud_GPU_30.job + # bsub 30.lsf + status: watch bjobs diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_1.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_1.job index 7e97ed6f..4c0e0d61 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_1.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_1.job @@ -3,7 +3,7 @@ #BSUB -nnodes 1 #BSUB -P GEN150_bench #BSUB -o cloud.o%J -#BSUB -J cloudJobx +#BSUB -J cloudmask_GPU_1 mkdir -p outputs-1 diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_12.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_12.job index 61bfaea9..6a853fa5 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_12.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_12.job @@ -3,7 +3,7 @@ #BSUB -nnodes 2 #BSUB -P GEN150_bench #BSUB -o cloud_GPU_12.o%J -#BSUB -J cloudJobx_GPU_12 +#BSUB -J cloudmask_GPU_12 mkdir -p outputs-12 diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_18.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_18.job index 603cbc53..6e06e73b 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_18.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_18.job @@ -3,7 +3,7 @@ #BSUB -nnodes 3 #BSUB -P GEN150_bench #BSUB -o cloud_GPU_18.o%J -#BSUB -J cloudJobx_GPU_18 +#BSUB -J cloudmask_GPU_18 mkdir -p outputs-18 diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_2.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_2.job index a36ec69c..6b116091 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_2.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_2.job @@ -3,12 +3,14 @@ #BSUB -nnodes 1 #BSUB -P GEN150_bench #BSUB -o cloud_GPU_2.o%J -#BSUB -J cloudJobx_GPU_2 +#BSUB -J cloudmask_GPU_2 +source $WORKDIR/ENV3/bin/activate mkdir -p outputs-2 # Load modules +module purge module load open-ce # Install libraries @@ -16,6 +18,7 @@ pip install scikit-learn pip install h5py pip install pyyaml pip install cloudmesh-common +pip install tensorflow #This runs on many nodes echo "Hostname: " diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_24.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_24.job index 32e9dc57..2ead56fd 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_24.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_24.job @@ -3,7 +3,7 @@ #BSUB -nnodes 4 #BSUB -P GEN150_bench #BSUB -o cloud_GPU_24.o%J -#BSUB -J cloudJobx_GPU_24 +#BSUB -J cloudmask_GPU_24 mkdir -p outputs-24 diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_30.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_30.job index da8c4b43..f3536449 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_30.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_30.job @@ -3,7 +3,7 @@ #BSUB -nnodes 30 #BSUB -P GEN150_bench #BSUB -o cloud_GPU_30.o%J -#BSUB -J cloudJobx_GPU_30 +#BSUB -J cloudmask_GPU_30 mkdir -p outputs-30 diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_4.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_4.job index 42b0ac0f..d21ee60b 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_4.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_4.job @@ -3,7 +3,7 @@ #BSUB -nnodes 1 #BSUB -P GEN150_bench #BSUB -o cloud_GPU_4.o%J -#BSUB -J cloudJobx_GPU_4 +#BSUB -J cloudmask_GPU_4 mkdir -p outputs-4 diff --git a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_6.job b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_6.job index ebdc02ad..2062933f 100644 --- a/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_6.job +++ b/benchmarks/cloudmask/target/summit-no-sciml/cloud_GPU_6.job @@ -4,7 +4,7 @@ #BSUB -nnodes 1 #BSUB -P GEN150_bench #BSUB -o cloud_GPU_6.o%J -#BSUB -J cloudJobx_GPU_6 +#BSUB -J cloudmask_GPU_6 mkdir -p outputs-6