Commit 660ced96 authored by Wang, Xiao's avatar Wang, Xiao
Browse files

pubmed crusher pretraining run script

parent c0ca8ce8
Loading
Loading
Loading
Loading
+178 −0
Original line number Diff line number Diff line
#!/bin/bash

#SBATCH -A MED106_crusher
#SBATCH -N 2
#SBATCH -t 00:20:00
#SBATCH -J crusher
#SBATCH -o %x-%j.out
#SBATCH -p batch

set +x

source /gpfs/alpine/med106/world-shared/irl1/crusher/miniconda/etc/profile.d/conda.sh
conda activate /ccs/proj/med106/working_clone


export PATH=/gpfs/alpine/world-shared/med106/gounley1/crusher/pdsh/src/pdsh:$PATH

module load PrgEnv-gnu
module load gcc/10.3.0
module load rocm/5.1.0
export ROCM_HOME=/opt/rocm-5.1.0
#export NCCL_SOCKET_IFNAME=hsn
export CC=gcc
export CXX=g++

#HOSTS=.hosts-job$SLURM_JOB_ID
#HOSTFILE=hostfile.txt
#srun hostname > $HOSTS
#sed 's/$/ slots=8/' $HOSTS > $HOSTFILE

ranks_per_node=8
gpus_per_rank=$((8/$ranks_per_node))
ranks_total=$(($ranks_per_node*$SLURM_JOB_NUM_NODES))
#export LD_PRELOAD="/opt/cray/pe/gcc/11.2.0/snos/lib64/libstdc++.so.6 /gpfs/alpine/world-shared/bip214/rocm_smi_lib/build/rocm_smi/librocm_smi64.so"
export LD_PRELOAD="/opt/cray/pe/gcc/11.2.0/snos/lib64/libstdc++.so.6.0.29 /gpfs/alpine/world-shared/bip214/rocm_smi_lib/build/rocm_smi/librocm_smi64.so"
export LD_PRELOAD="${LD_PRELOAD} ${CRAY_MPICH_ROOTDIR}/gtl/lib/libmpi_gtl_hsa.so"


#module load open-ce
#conda deactivate
#conda activate /gpfs/alpine/med106/world-shared/irl1/rhel8/myt_py1.11

#conda activate /gpfs/alpine/med106/world-shared/irl1/rhel8/myt_py1.8
#conda activate /gpfs/alpine/med106/world-shared/irl1/rhel8/megaCE
module list

export OMP_NUM_THREADS=1
#export PYTHONPATH=$PYTHONPATH:/gpfs/alpine/med106/world-shared/irl1/rhel8/fork-megatron/megatron/fused_kernels
#export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/gpfs/alpine/med106/world-shared/irl1/rhel8/mytorch/lib/python3.8/site-packages/torch/lib
#export PATH=$PATH:/gpfs/alpine/med106/world-shared/irl1/rhel8/mytorch/lib/python3.8/site-packages/torch/include

DATASET_0="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm000_abstract_sentence"
DATASET_1="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm001_abstract_sentence"
DATASET_2="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm002_abstract_sentence"
DATASET_3="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm003_abstract_sentence"
DATASET_4="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm004_abstract_sentence"
DATASET_5="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm005_abstract_sentence"
DATASET_6="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm006_abstract_sentence"
DATASET_7="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm007_abstract_sentence"
DATASET_8="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm008_abstract_sentence"
DATASET_9="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm009_abstract_sentence"
DATASET_10="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm010_abstract_sentence"
DATASET_11="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm011_abstract_sentence"
DATASET_12="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm012_abstract_sentence"
DATASET_13="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm013_abstract_sentence"
DATASET_14="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm014_abstract_sentence"
DATASET_15="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm015_abstract_sentence"
DATASET_16="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm016_abstract_sentence"
DATASET_17="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm017_abstract_sentence"
DATASET_18="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm018_abstract_sentence"
DATASET_19="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm019_abstract_sentence"
DATASET_20="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm020_abstract_sentence"
DATASET_21="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm021_abstract_sentence"
DATASET_22="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm022_abstract_sentence"
DATASET_23="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm023_abstract_sentence"
DATASET_24="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm024_abstract_sentence"
DATASET_25="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm025_abstract_sentence"
DATASET_26="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm026_abstract_sentence"
DATASET_27="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm027_abstract_sentence"
DATASET_28="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm028_abstract_sentence"
DATASET_29="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm029_abstract_sentence"
DATASET_30="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm030_abstract_sentence"
DATASET_31="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm031_abstract_sentence"
DATASET_32="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm032_abstract_sentence"
DATASET_33="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm033_abstract_sentence"
DATASET_34="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm034_abstract_sentence"
DATASET_35="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm035_abstract_sentence"
DATASET_36="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm036_abstract_sentence"
DATASET_37="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm037_abstract_sentence"
DATASET_38="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm038_abstract_sentence"
DATASET_39="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm039_abstract_sentence"
DATASET_40="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm040_abstract_sentence"
DATASET_41="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm041_abstract_sentence"
DATASET_42="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm042_abstract_sentence"
DATASET_43="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm043_abstract_sentence"
DATASET_44="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm044_abstract_sentence"
DATASET_45="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm045_abstract_sentence"
DATASET_46="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm046_abstract_sentence"
DATASET_47="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm047_abstract_sentence"
DATASET_48="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm048_abstract_sentence"
DATASET_49="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm049_abstract_sentence"
DATASET_50="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm050_abstract_sentence"
DATASET_51="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm051_abstract_sentence"
DATASET_52="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm052_abstract_sentence"
DATASET_53="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm053_abstract_sentence"
DATASET_54="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm054_abstract_sentence"
DATASET_55="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm055_abstract_sentence"
DATASET_56="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm056_abstract_sentence"
DATASET_57="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm057_abstract_sentence"
DATASET_58="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm058_abstract_sentence"
DATASET_59="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm059_abstract_sentence"
DATASET_60="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm060_abstract_sentence"
DATASET_61="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm061_abstract_sentence"
DATASET_62="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm062_abstract_sentence"
DATASET_63="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm063_abstract_sentence"
DATASET_64="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm064_abstract_sentence"
DATASET_65="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm065_abstract_sentence"
DATASET_66="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm066_abstract_sentence"
DATASET_67="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm067_abstract_sentence"
DATASET_68="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm068_abstract_sentence"
DATASET_69="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm069_abstract_sentence"
DATASET_70="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm070_abstract_sentence"
DATASET_71="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm071_abstract_sentence"
DATASET_72="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm072_abstract_sentence"
DATASET_73="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm073_abstract_sentence"
DATASET_74="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm074_abstract_sentence"
DATASET_75="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm075_abstract_sentence"
DATASET_76="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm076_abstract_sentence"
DATASET_77="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm077_abstract_sentence"
DATASET_78="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm078_abstract_sentence"
DATASET_79="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm079_abstract_sentence"
DATASET_80="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm080_abstract_sentence"
DATASET_81="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm081_abstract_sentence"
DATASET_82="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm082_abstract_sentence"
DATASET_83="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm083_abstract_sentence"
DATASET_84="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm084_abstract_sentence"
DATASET_85="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm085_abstract_sentence"
DATASET_86="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm086_abstract_sentence"
DATASET_87="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm087_abstract_sentence"
DATASET_88="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm088_abstract_sentence"
DATASET_89="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm089_abstract_sentence"
DATASET_90="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm090_abstract_sentence"
DATASET_91="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm091_abstract_sentence"
DATASET_92="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm092_abstract_sentence"
DATASET_93="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm093_abstract_sentence"
DATASET_94="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm094_abstract_sentence"
DATASET_95="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm095_abstract_sentence"
DATASET_96="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm096_abstract_sentence"
DATASET_97="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm097_abstract_sentence"
DATASET_98="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm098_abstract_sentence"
DATASET_99="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm099_abstract_sentence"
DATASET_100="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm100_abstract_sentence"
DATASET_101="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm101_abstract_sentence"
DATASET_102="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm102_abstract_sentence"
DATASET_103="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm103_abstract_sentence"
DATASET_104="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm104_abstract_sentence"
DATASET_105="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm105_abstract_sentence"
DATASET_106="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm106_abstract_sentence"
DATASET_107="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm107_abstract_sentence"
DATASET_108="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm108_abstract_sentence"
DATASET_109="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm109_abstract_sentence"
DATASET_110="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm110_abstract_sentence"
DATASET_111="/gpfs/alpine/med106/proj-shared/xf9/pubmed_data/preprocessed/pm111_abstract_sentence"




DATASET="1.0 ${DATASET_0} 1.0 ${DATASET_1} 1.0 ${DATASET_2} 1.0 ${DATASET_3} 1.0 ${DATASET_4} 1.0 ${DATASET_5} 1.0 ${DATASET_6} 1.0 ${DATASET_7} 1.0 ${DATASET_8} 1.0 ${DATASET_9} 1.0 ${DATASET_10} 1.0 ${DATASET_11} 1.0 ${DATASET_12} 1.0 ${DATASET_13} 1.0 ${DATASET_14} 1.0 ${DATASET_15} 1.0 ${DATASET_16} 1.0 ${DATASET_17} 1.0 ${DATASET_18} 1.0 ${DATASET_19} 1.0 ${DATASET_20} 1.0 ${DATASET_21} 1.0 ${DATASET_22} 1.0 ${DATASET_23} 1.0 ${DATASET_24} 1.0 ${DATASET_25} 1.0 ${DATASET_26} 1.0 ${DATASET_27} 1.0 ${DATASET_28} 1.0 ${DATASET_29} 1.0 ${DATASET_30} 1.0 ${DATASET_31} 1.0 ${DATASET_32} 1.0 ${DATASET_33} 1.0 ${DATASET_34} 1.0 ${DATASET_35} 1.0 ${DATASET_36} 1.0 ${DATASET_37} 1.0 ${DATASET_38} 1.0 ${DATASET_39} 1.0 ${DATASET_40} 1.0 ${DATASET_41} 1.0 ${DATASET_42} 1.0 ${DATASET_43} 1.0 ${DATASET_44} 1.0 ${DATASET_45} 1.0 ${DATASET_46} 1.0 ${DATASET_47} 1.0 ${DATASET_48} 1.0 ${DATASET_49} 1.0 ${DATASET_50} 1.0 ${DATASET_51} 1.0 ${DATASET_52} 1.0 ${DATASET_53} 1.0 ${DATASET_54} 1.0 ${DATASET_55} 1.0 ${DATASET_56} 1.0 ${DATASET_57} 1.0 ${DATASET_58} 1.0 ${DATASET_59} 1.0 ${DATASET_60} 1.0 ${DATASET_61} 1.0 ${DATASET_62} 1.0 ${DATASET_63} 1.0 ${DATASET_64} 1.0 ${DATASET_65} 1.0 ${DATASET_66} 1.0 ${DATASET_67} 1.0 ${DATASET_68} 1.0 ${DATASET_69} 1.0 ${DATASET_70} 1.0 ${DATASET_71} 1.0 ${DATASET_72} 1.0 ${DATASET_73} 1.0 ${DATASET_74} 1.0 ${DATASET_75} 1.0 ${DATASET_76} 1.0 ${DATASET_77} 1.0 ${DATASET_78} 1.0 ${DATASET_79} 1.0 ${DATASET_80} 1.0 ${DATASET_81} 1.0 ${DATASET_82} 1.0 ${DATASET_83} 1.0 ${DATASET_84} 1.0 ${DATASET_85} 1.0 ${DATASET_86} 1.0 ${DATASET_87} 1.0 ${DATASET_88} 1.0 ${DATASET_89} 1.0 ${DATASET_90} 1.0 ${DATASET_91} 1.0 ${DATASET_92} 1.0 ${DATASET_93} 1.0 ${DATASET_94} 1.0 ${DATASET_95} 1.0 ${DATASET_96} 1.0 ${DATASET_97} 1.0 ${DATASET_98} 1.0 ${DATASET_99} 1.0 ${DATASET_100} 1.0 ${DATASET_101} 1.0 ${DATASET_102} 1.0 ${DATASET_103} 1.0 ${DATASET_104} 1.0 ${DATASET_105} 1.0 ${DATASET_106} 1.0 ${DATASET_107} 1.0 ${DATASET_108} 1.0 ${DATASET_109} 1.0 ${DATASET_110} 1.0 ${DATASET_111}"
#DATASET="1.0 ${DATASET_0} 1.0 ${DATASET_1} 1.0 ${DATASET_2}"

export VOCAB_FILE=/gpfs/alpine/world-shared/med106/g8o/pubmed_bert-vocab.txt
export CHECKPOINT_PATH=/gpfs/alpine/med106/scratch/xf9/Megatron-LM/2N6D2T1P


srun -u -n $ranks_total -c 8 --gpus-per-task=$gpus_per_rank --gpu-bind=closest bash -c "
source export_DDP_vars.sh
/ccs/proj/med106/working_clone/bin/python pretrain_bert.py --tensor-model-parallel-size 2 --pipeline-model-parallel-size 2 --num-layers 24 --hidden-size 1024 --num-attention-heads 16 --micro-batch-size 2 --seq-length 512 --max-position-embeddings 512 --train-iters 2000 --lr-decay-iters 990 --save $CHECKPOINT_PATH --data-path $DATASET --vocab-file $VOCAB_FILE --data-impl mmap --split 949,50,1 --distributed-backend nccl --lr 0.0001 --min-lr 0.00001 --lr-decay-style linear --lr-warmup-fraction .01 --weight-decay 1e-2 --clip-grad 1.0 --log-interval 100 --save-interval 10000 --eval-interval 1000 --eval-iters 10 --HIP 1 --fp16"