stage3-draw-retrain_ticket.sh

#!/bin/bash
export CUDA_DEVICE_ORDER=PCI_BUS_ID
export CUDA_VISIBLE_DEVICES=0

log_file='draw-retrain_ticket.log'
result_file='retrain_ticket.csv'  # attack results
ckpt='./save_models/draw-retrain_ticket/'  # draw and retrain tickets saving path
lr=2e-5
epoch=10  # larger epoch is better

# IMDB
ticket_path='./your_search-ticket_path' # search ticket path
for epoch in 19 17 15 13
do
  for sparsity in 0.2 0.3 0.4
  do
    masked_model_path=${model_path}${epoch} # retrain on tickets with different searching epochs
    python draw_retrain_ticket.py \
    --masked_model_path ${ticket_path} \
    --model_name bert-base-uncased \
    --lr $lr \
    --max_seq_length 256 \
    --dataset_name imdb \
    --ckpt_dir ${ckpt} \
    --result_file ${result_file} \
    --sparsity $sparsity \
    --num_examples 100 \
    --bsz 32 \
    --num_labels 2 \
    --epochs $epoch >> ${log_file}
done
done

# AGNEWS
ticket_path='./your_search-ticket_path' # search ticket path
for epoch in 19 17 15 13
do
  for sparsity in 0.2 0.3 0.4
  do
    masked_model_path=${model_path}${epoch} # retrain on tickets with different searching epochs
    python draw_retrain_ticket.py \
    --masked_model_path ${ticket_path} \
    --model_name bert-base-uncased \
    --lr $lr \
    --max_seq_length 256 \
    --dataset_name ag_news \
    --ckpt_dir ${ckpt} \
    --result_file ${result_file} \
    --sparsity $sparsity \
    --num_examples 200 \
    --bsz 32 \
    --num_labels 4 \
    --epochs $epoch >> ${log_file}
done
done

# SST-2
ticket_path='./your_search-ticket_path' # search ticket path
for epoch in 19 17 15 13
do
  for sparsity in 0.2 0.3 0.4
  do
    masked_model_path=${model_path}${epoch} # retrain on tickets with different searching epochs
    python draw_retrain_ticket.py \
    --masked_model_path ${ticket_path} \
    --model_name bert-base-uncased \
    --lr $lr \
    --max_seq_length 128 \
    --dataset_name glue \
    --task_name sst2 \
    --ckpt_dir ${ckpt} \
    --result_file ${result_file} \
    --sparsity $sparsity \
    --num_examples 872 \
    --bsz 32 \
    --num_labels 2 \
    --epochs $epoch >> ${log_file}
done
done