Adding CoriGPU run files and modifying conf.yaml

Techercise · Techercise · commit 0b59a7a8bf57 · 2021-06-23T06:43:56.000-07:00
diff --git a/examples/conf.yaml b/examples/conf.yaml
@@ -4,7 +4,7 @@
 
 # will do stuff in fs_path / [username] / signal_data | shot_lists | processed shots, etc.
 
-fs_path: '/tigress'
+fs_path: '/global/cscratch1/sd/'
 target: 'hinge' # 'maxhinge' # 'maxhinge' # 'binary' # 'hinge'
 num_gpus: 4  # per node
 paths:
@@ -127,7 +127,7 @@ training:
   num_shots_at_once: 200
   # large number = maximum number of epochs.
   # Early stopping will occur if loss does not decrease, after some patience # of epochs
-  num_epochs: 1000
+  num_epochs: 50
   use_mock_data: False
   data_parallel: False
   hyperparam_tuning: False
@@ -136,8 +136,8 @@ training:
   num_batches_minimum: 20 # minimum number of batches per epoch
   ranking_difficulty_fac: 1.0 # how much to upweight incorrectly classified shots during training
   timeline_prof: False
-  step_limit: 50
-  no_validation: True
+  step_limit: 0
+  no_validation: False
 callbacks:
   list: ['earlystop']
   metrics: ['val_loss','val_roc','train_loss']
diff --git a/examples/corigpu_1GPU_slurm.cmd b/examples/corigpu_1GPU_slurm.cmd
@@ -0,0 +1,15 @@
+#!/bin/bash
+#SBATCH -C gpu
+#SBATCH -t 01:30:00
+#SBATCH -G 1
+#SBATCH -c 4
+#SBATCH --exclusive
+
+# rm /global/cscratch1/sd/$USER/model_checkpoints/*
+# rm /global/cscratch1/sd/$USER/results/*
+# rm /global/cscratch1/sd/$USER/csv_logs/*
+# rm /global/cscratch1/sd/$USER/Graph/*
+# rm /global/cscratch1/sd/$USER/normalization/*
+
+export OMPI_MCA_btl="tcp,self,vader"
+srun python mpi_learn.py
diff --git a/examples/corigpu_4GPU_slurm.cmd b/examples/corigpu_4GPU_slurm.cmd
@@ -0,0 +1,19 @@
+#!/bin/bash
+#SBATCH -C gpu
+#SBATCH -t 02:00:00
+#SBATCH -N 4
+#SBATCH -G 4
+#SBATCH --ntasks-per-node=4
+#SBATCH --ntasks-per-socket=2
+#SBATCH -c 4
+#SBATCH --mem-per-cpu=0
+#SBATCH --exclusive
+
+# rm /global/cscratch1/sd/$USER/model_checkpoints/*
+# rm /global/cscratch1/sd/$USER/results/*
+# rm /global/cscratch1/sd/$USER/csv_logs/*
+# rm /global/cscratch1/sd/$USER/Graph/*
+# rm /global/cscratch1/sd/$USER/normalization/*
+
+export OMPI_MCA_btl="tcp,self,vader"
+srun python mpi_learn.py
