clean up

ascent/README.md

-# imagenet_tutorial
-
-This code is from [NVIDIA-DeepLearningExamples](https://github.com/NVIDIA/DeepLearningExamples) with some modifications.
\ No newline at end of file

ascent/export_DDP_envvars.sh

-export RANK=$OMPI_COMM_WORLD_RANK
-export LOCAL_RANK=$OMPI_COMM_WORLD_LOCAL_RANK
-export WORLD_SIZE=$OMPI_COMM_WORLD_SIZE
-export MASTER_ADDR=$(cat $LSB_DJOB_HOSTFILE | sort | uniq | grep -v batch | grep -v login | head -1)
-export MASTER_PORT=29500 # default from torch launcher

ascent/sub_ibmwml.lsf

-#!/bin/bash
-# Begin LSF directives
-#BSUB -P stf011
-#BSUB -J sc21
-#BSUB -o logs/sc21.o%J
-#BSUB -W 0:30
-#BSUB -nnodes 1
-#BSUB -alloc_flags "nvme smt4"
-####BSUB -N
-# End LSF directives and begin shell commands
-
-nnodes=$(cat ${LSB_DJOB_HOSTFILE} | sort | uniq | grep -v login | grep -v batch | wc -l)
-DATA_DIR=/gpfs/wolf/gen166/proj-shared/atsaris/imagenet/data/
-LOG_DIR=logs/
-
-#source /gpfs/wolf/stf011/proj-shared/28t/env/open-ce/activate.sh
-source /gpfs/wolf/stf011/proj-shared/28t/env/ibm-wml/activate.sh
-
-jsrun --smpiargs="-disable_gpu_hooks" -n${nnodes} -a1 -c42 -g1 -r1 \
-    --bind=proportional-packed:7 --launch_distribution=packed \
-    bash -c "\
-    source export_DDP_envvars.sh && \
-    python -u ../imagenet/main.py \
-    --arch resnet50 \
-    -j 28 \
-    -p 10 \
-    -b 128\
-    --training-only \
-    --raport-file ${LOG_DIR}/synthetic.ibmwml.1GPU.json \
-    --epochs 1 \
-    --prof 100 \
-    --no-checkpoints \
-    --data-backend sythetic \
-    --amp \
-    --memory-format nhwc \
-    ${DATA_DIR}
-    "

ascent/sub_opence.lsf

-#!/bin/bash
-# Begin LSF directives
-#BSUB -P stf011
-#BSUB -J sc21
-#BSUB -o logs/sc21.o%J
-#BSUB -W 0:30
-#BSUB -nnodes 1
-#BSUB -alloc_flags "nvme smt4"
-####BSUB -N
-# End LSF directives and begin shell commands
-
-nnodes=$(cat ${LSB_DJOB_HOSTFILE} | sort | uniq | grep -v login | grep -v batch | wc -l)
-DATA_DIR=/gpfs/wolf/gen166/proj-shared/atsaris/imagenet/data/
-LOG_DIR=logs/
-
-source /gpfs/wolf/stf011/proj-shared/28t/env/open-ce/activate.sh
-#source /gpfs/wolf/stf011/proj-shared/28t/env/ibm-wml/activate.sh
-
-jsrun --smpiargs="-disable_gpu_hooks" -n${nnodes} -a1 -c42 -g1 -r1 \
-    --bind=proportional-packed:7 --launch_distribution=packed \
-    bash -c "\
-    source export_DDP_envvars.sh && \
-    python -u ../imagenet/main.py \
-    --arch resnet50 \
-    -j 28 \
-    -p 10 \
-    -b 128\
-    --training-only \
-    --raport-file ${LOG_DIR}/pytorch.opence.1GPU.json \
-    --epochs 1 \
-    --prof 100 \
-    --no-checkpoints \
-    --data-backend pytorch \
-    --amp \
-    --memory-format nhwc \
-    ${DATA_DIR}
-    "