step3_test.sh

num=1


# #dev
# if true; then
# validation_file=/data4/cqdong/my_align/zh2en/zh2en/TsinghuaAlignmentEvalSet/v1/devset/dev.ch2en
# validation_gold_file=/data4/cqdong/my_align/zh2en/zh2en/TsinghuaAlignmentEvalSet/v1/devset/dev.wa.formal.align
# cache_data=/data4/cqdong/my_align/cache/dev_ch2en40k.pt
# output_dir=/data4/cqdong/my_align/output/ch2en40dev/
# fi
# test
if true; then
validation_file=/data4/cqdong/my_align/zh2en/zh2en/TsinghuaAlignmentEvalSet/v1/tstset/tst.ch2en
validation_gold_file=/data4/cqdong/my_align/zh2en/zh2en/TsinghuaAlignmentEvalSet/v1/tstset/tst.wa.formal.align
cache_data=/data4/cqdong/my_align/cache/tst_de2en40k.pt
output_dir=/data4/cqdong/my_align/output/ch2en40tst/
fi


for check_num in {4..4};do
for layers in {8..8};do
CUDA_VISIBLE_DEVICES=$num python /data4/cqdong/my_align/src/align_extract.py \
    --validation_file  $validation_file \
    --validation_gold_file $validation_gold_file \
    --model_name_or_path /data4/cqdong/my_align/checkpoints_zh_en_dot_bert_1e-5_gold/model_${check_num}\.pt\
    --tokenizer_name /data4/jpzhang/lexicon_induction/pretrain_model/bert-base-multilingual-cased\
    --config_name /data4/jpzhang/lexicon_induction/pretrain_model/bert-base-multilingual-cased\
    --cache_data $cache_data \
    --overwrite_cache \
    --output_dir $output_dir \
    --block_size 512 \
    --encoder_layers $layers \
    --per_device_eval_batch_size 16 \
    --model_type bert \
    --do_eval 
    
# test AER
python /data4/cqdong/Mask-Align/alignment-scripts/scripts/aer.py\
    $validation_gold_file\
    $output_dir/align.txt\
    --oneRef --fAlpha 0.5 >>zh2en.out
    echo $layers>>zh2en.out
done
done