conda create -n csqe python=3.9
conda activate csqe
conda install -c conda-forge faiss-gpu openjdk=11 maven
pip install transformers==4.20.0
pip install openai pandas matplotlib gpustat ipython jupyter datasets accelerate sklearn tensorboard nltk pytrec_eval prettytable gradio
pip install pyserini
pip install setuptools==59.5.0
python -m spacy download en_core_web_sm
python3 csqe.py \
--expansion_method keqe/csqe \
--threads 16 --batch-size 128 --index msmarco-v1-passage \
--topics dl19-passage --answer_key contents \
--bm25 --disable_bm25_param --max_demo_len 128 \
--openai_api_key <API_KEY> --trec_python_path <PYTHON_PATH> \
--output_dir <OUTPUT_DIR>
python3 csqe.py \
--expansion_method keqe/csqe \
--threads 16 --batch-size 128 --index msmarco-v1-passage \
--topics dl20 --qrels dl20-passage --answer_key contents \
--bm25 --disable_bm25_param --max_demo_len 128 \
--openai_api_key <API_KEY> --trec_python_path <PYTHON_PATH> \
--output_dir <OUTPUT_DIR>
DATASET="trec-covid"
# can also be "scifact", "arguana", "fiqa", "dbpedia-entity", "trec-news"
python3 csqe.py \
--expansion_method keqe/csqe \
--threads 16 --batch-size 128 --index beir-v1.0.0-${DATASET}.flat --topics beir-v1.0.0-${DATASET}-test \
--answer_key "title|text" --metric_beir \
--bm25 --disable_bm25_param --max_demo_len 128 \
--openai_api_key <API_KEY> --trec_python_path <API_KEY> \
--output_dir <OUTPUT_DIR>
When using csqe, please add --gen_num 2 to constrain both keqe and corpus-originated expansion only generate two generations.
For "arguana", please add --remove-query.