wanchichen's picture
Add files using upload-large-folder tool
3049ae7 verified
#!/bin/bash
cd /mnt/home/williamchen/espnet/egs2/owsm_v3.1/s2t1
. ./path.sh
( echo '#' Running on `hostname`
echo '#' Started at `date`
set | grep SLURM | while read line; do echo "# $line"; done
echo -n '# '; cat <<EOF
srun --export=ALL python3 -m espnet2.bin.s2t_train --use_preprocessor true --bpemodel data/token_list/bpe_unigram50000/bpe.model --token_type bpe --token_list data/token_list/bpe_unigram50000/tokens.txt --non_linguistic_symbols none --cleaner none --g2p none --valid_data_path_and_name_and_type dump/raw/dev_v3/wav.scp,speech,kaldi_ark --valid_shape_file exp_owsm/s2t_stats_raw_bpe50000/valid/speech_shape --resume true --fold_length 80000 --output_dir exp_owsm/s2t_train_9b_ds_raw_bpe50000 --config conf/train_9b_ds.yaml --frontend_conf fs=16k --normalize=global_mvn --normalize_conf stats_file=exp_owsm/s2t_stats_raw_bpe50000/train/feats_stats.npz --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/wav.scp,speech,kaldi_ark --train_shape_file exp_owsm/s2t_stats_raw_bpe50000/splits8/text_shape --fold_length 150 --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/text.prev,text_prev,text --fold_length 150 --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/text.ctc,text_ctc,text --fold_length 150 --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/text,text,text --multiple_iterator true --valid_data_path_and_name_and_type dump/raw/dev_v3/text.prev,text_prev,text --valid_data_path_and_name_and_type dump/raw/dev_v3/text.ctc,text_ctc,text --valid_data_path_and_name_and_type dump/raw/dev_v3/text,text,text --ngpu 8 --multiprocessing_distributed true --dist_launcher slurm --dist_init_method file:///mnt/home/williamchen/espnet/egs2/owsm_v3.1/s2t1/exp_owsm/s2t_train_9b_ds_raw_bpe50000/.dist_init_6ad39c72-e06d-4dd3-9418-3927b0b6d780
EOF
) >exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log
if [ "$CUDA_VISIBLE_DEVICES" == "NoDevFiles" ]; then
( echo CUDA_VISIBLE_DEVICES set to NoDevFiles, unsetting it...
)>>exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log
unset CUDA_VISIBLE_DEVICES
fi
time1=`date +"%s"`
( srun --export=ALL python3 -m espnet2.bin.s2t_train --use_preprocessor true --bpemodel data/token_list/bpe_unigram50000/bpe.model --token_type bpe --token_list data/token_list/bpe_unigram50000/tokens.txt --non_linguistic_symbols none --cleaner none --g2p none --valid_data_path_and_name_and_type dump/raw/dev_v3/wav.scp,speech,kaldi_ark --valid_shape_file exp_owsm/s2t_stats_raw_bpe50000/valid/speech_shape --resume true --fold_length 80000 --output_dir exp_owsm/s2t_train_9b_ds_raw_bpe50000 --config conf/train_9b_ds.yaml --frontend_conf fs=16k --normalize=global_mvn --normalize_conf stats_file=exp_owsm/s2t_stats_raw_bpe50000/train/feats_stats.npz --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/wav.scp,speech,kaldi_ark --train_shape_file exp_owsm/s2t_stats_raw_bpe50000/splits8/text_shape --fold_length 150 --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/text.prev,text_prev,text --fold_length 150 --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/text.ctc,text_ctc,text --fold_length 150 --train_data_path_and_name_and_type exp_owsm/s2t_stats_raw_bpe50000/splits8/text,text,text --multiple_iterator true --valid_data_path_and_name_and_type dump/raw/dev_v3/text.prev,text_prev,text --valid_data_path_and_name_and_type dump/raw/dev_v3/text.ctc,text_ctc,text --valid_data_path_and_name_and_type dump/raw/dev_v3/text,text,text --ngpu 8 --multiprocessing_distributed true --dist_launcher slurm --dist_init_method file:///mnt/home/williamchen/espnet/egs2/owsm_v3.1/s2t1/exp_owsm/s2t_train_9b_ds_raw_bpe50000/.dist_init_6ad39c72-e06d-4dd3-9418-3927b0b6d780 ) &>>exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log
ret=$?
sync || true
time2=`date +"%s"`
echo '#' Accounting: begin_time=$time1 >>exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log
echo '#' Accounting: end_time=$time2 >>exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log
echo '#' Accounting: time=$(($time2-$time1)) threads=1 >>exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log
echo '#' Finished at `date` with status $ret >>exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log
[ $ret -eq 137 ] && exit 100;
touch exp_owsm/s2t_train_9b_ds_raw_bpe50000/q/done.2830290
exit $[$ret ? 1 : 0]
## submitted with:
# sbatch --export=PATH --job-name exp_owsm/s2t_train_9b_ds_raw_bpe50000/train.log --nodes 3 --cpus-per-task 8 -p gpu --gres=gpu:8 -c 8 --gpus-per-node=8 --partition p2 --cpus-per-task=128 --mem=2000G --exclusive --open-mode=append -e exp_owsm/s2t_train_9b_ds_raw_bpe50000/q/train.log -o exp_owsm/s2t_train_9b_ds_raw_bpe50000/q/train.log /mnt/home/williamchen/espnet/egs2/owsm_v3.1/s2t1/exp_owsm/s2t_train_9b_ds_raw_bpe50000/q/train.sh >>exp_owsm/s2t_train_9b_ds_raw_bpe50000/q/train.log 2>&1