run_conll03_spanner.sh

export PYTHONPATH="$PWD"

DATA_DIR="data/conll03"
PRETRAINED="bert-large-uncased"
BERT_DIR="/home/jlfu/Projects/"${PRETRAINED}




dataname=conll03
n_class=5
BERT_DROPOUT=0.2
MODEL_DROPOUT=0.2
LR=1e-5
MAXLEN=128
MAXNORM=1.0
batchSize=10
max_spanLen=4
tokenLen_emb_dim=50
spanLen_emb_dim=100
morph_emb_dim=100


use_prune=True
use_spanLen=True
use_morph=True
use_span_weight=True
neg_span_weight=0.5
gpus="7,"



#max_epochs=30
max_epochs=1
modelName="spanner_"${PRETRAINED}_spMLen${max_span_len}_usePrune${use_prune}_useSpLen${use_spanLen}_useSpMorph${use_morph}_SpWt${use_span_weight}_value${neg_span_weight}
idtest=${dataname}_${modelName}
param_name=epoch${max_epochs}_batchsize${batchSize}_lr${LR}_maxlen${MAXLEN}

OUTPUT_DIR="/home/jlfu/spanner/train_logs/$dataname/${modelName}"
#mkdir -p $OUTPUT_DIR

nohup python trainer.py \
--dataname $dataname \
--data_dir $DATA_DIR \
--bert_config_dir $BERT_DIR \
--bert_max_length $MAXLEN \
--batch_size $batchSize \
--gpus=$gpus \
--precision=16 \
--progress_bar_refresh_rate 1 \
--lr $LR \
--distributed_backend=dp \
--val_check_interval 0.25 \
--accumulate_grad_batches 1 \
--default_root_dir $OUTPUT_DIR \
--model_dropout $MODEL_DROPOUT \
--bert_dropout $BERT_DROPOUT \
--max_epochs $max_epochs \
--n_class $n_class \
--max_spanLen $max_spanLen \
--tokenLen_emb_dim $tokenLen_emb_dim \
--modelName $modelName \
--spanLen_emb_dim $spanLen_emb_dim \
--morph_emb_dim $morph_emb_dim \
--use_prune $use_prune \
--use_spanLen $use_spanLen \
--use_morph $use_morph \
--use_span_weight $use_span_weight \
--neg_span_weight $neg_span_weight \
--param_name $param_name \
--gradient_clip_val $MAXNORM \
--optimizer "adamw" >train_logs/${idtest}_epoch${max_epochs}_0606.txt &