Experiment 🧪

lang=all

lr=5e-5

batch_size=4

source_length=512

data_dir=dataset_fine_grain

label_num=6

output_dir=model_${label_num}/$lang

train_file=$data_dir/$lang/train.json

dev_file=$data_dir/$lang/dev.json

test_file=$data_dir/$lang/test.json

epochs=2

debug_size=10

delete_weight=4

replace_weight=4

insert_weight=4

block_split_weight=12

select_method=bm25

model_type=codet5

pretrained_model=salesforce/codet5-large

load_locator_model_path=model/$lang/checkpoint-last/pytorch_model.bin

python run.py --lang $lang --do_test --do_train --do_eval --model_type $model_type --model_name_or_path $pretrained_model --train_filename $train_file --dev_filename $dev_file --test_filename $test_file --output_dir $output_dir --max_source_length $source_length --locator_batch_size $batch_size --learning_rate $lr --num_train_epochs $epochs --delete_weight $delete_weight --replace_weight $replace_weight --insert_weight $insert_weight --block_split_weight $block_split_weight --select_method $select_method --label_num $label_num 

LANG=all

MODEL_TAG=Salesforce/codet5-base

GPU=0

DATA_NUM=-1

BS=4

LR=5e-5

SRC_LEN=512

TRG_LEN=128

PATIENCE=5

EPOCH=1

WARMUP=100

paper_name=new

label_num=6

MODEL_DIR=./model_${label_num}_${paper_name}

OUTPUT_DIR=${MODEL_DIR}/${LANG}

SUMMARY_DIR=.

DATA_DIR=dataset_fine_grain/${LANG}

RES_DIR=${OUTPUT_DIR}/prediction

LOG=${OUTPUT_DIR}/train.log

SELECT_METHOD=bm25

beam_size=10

mkdir -p ${OUTPUT_DIR}

mkdir -p ${RES_DIR}

CUDA_VISIBLE_DEVICES=1

python run.py --do_train --do_eval --do_test --model_type codet5 --data_num $DATA_NUM --num_train_epochs $EPOCH --warmup_steps $WARMUP --learning_rate $LR --patience $PATIENCE --tokenizer_name $MODEL_TAG --model_name_or_path $MODEL_TAG --data_dir $DATA_DIR --output_dir $OUTPUT_DIR  --summary_dir $SUMMARY_DIR --save_last_checkpoints --always_save_model --res_dir $RES_DIR --estimator_batch_size $BS --train_batch_size $BS --eval_batch_size $BS --max_source_length $SRC_LEN --max_target_length $TRG_LEN --select_method $SELECT_METHOD --label_num $label_num --beam_size $beam_size --paper_name $paper_name 

lr=5e-5

batch_size=16

source_length=512

data_dir=dataset

output_dir=model

train_file=$data_dir/train.json

dev_file=$data_dir/valid.json

test_file=$data_dir/test.json

epochs=2

debug_size=10

model_type=roberta

pretrained_model=microsoft/codebert-base

load_locator_model_path=model/checkpoint-last/pytorch_model.bin

python run.py \

--do_test --do_train --do_eval \

--model_type $model_type --model_name_or_path $pretrained_model \

--train_filename $train_file --dev_filename $dev_file --test_filename $test_file \

--output_dir $output_dir --dataset_dir $data_dir \

--max_source_length $source_length \

--batch_size $batch_size \

--learning_rate $lr --num_train_epochs $epochs \