run_biencoder_unixcoder_pl_fp16.sh

export CUDA_VISIBLE_DEVICES=3

pl=$1 # name of programming language

model_name=unixcoder

MODEL_DIR=tmp/bi_fp16_${model_name}_${pl}
EMBEDDINGS_DIR=tmp/bi_fp16_${model_name}_embeddings_${pl}
RESULT_DIR=tmp/bi_fp16_${model_name}_result_${pl}

mkdir -p $RESULT_DIR

python run_biencoder.py \
    --model_name_or_path ../models/unixcoder-base \
    --output_dir $MODEL_DIR \
    --data_file ../pls/qa.en.${pl}.json \
    --train_group_size 1 \
    --do_train \
    --per_device_train_batch_size 32 \
    --num_train_epochs 3 \
    --do_eval \
    --per_device_eval_batch_size 64 \
    --save_strategy epoch \
    --evaluation_strategy epoch \
    --overwrite_output_dir \
    --logging_steps 100 \
    --fp16


python run_biencoder.py \
    --model_name_or_path $MODEL_DIR \
    --output_dir $MODEL_DIR \
    --data_file ../pls/qa.en.${pl}.json \
    --do_predict \
    --prediction_save_path $EMBEDDINGS_DIR \
    --encode_corpus \
    --encode_query \
    --overwrite_output_dir \
    --fp16


python test.py \
    --query_reps_path $EMBEDDINGS_DIR/query_reps \
    --passage_reps_path $EMBEDDINGS_DIR/passage_reps \
    --qrels_file $EMBEDDINGS_DIR/qrels.test.tsv \
    --ranking_file  $EMBEDDINGS_DIR/test_ranking.txt \
    > ${RESULT_DIR}/result.log 2>&1