pip install -r requirements.txt

Run retrieval model (ColBert)

Download the corpus for retrieval from DPR

wget https://dl.fbaipublicfiles.com/dpr/wikipedia_split/psgs_w100.tsv.gz

Indext your data

python ./retrieval/ColBERT/index.py

Run the service for retrieval

python ./retrieval/ColBERT/server_retrieval.py

Retrieval for your dataset

python ./retrieval/retrieval_top_k_data.py

Primary experiment

/app/anaconda3/bin/deepspeed --num_gpus=1 --master_addr="127.0.0.0" --master_port=29560 comparing_benefit_and_detriment.py \
   --file_path wikitext103 \ # Your dataset path, such as wikitext103
   --model_name_or_path $Your_LLM_path \ # Your LLm path
   --seed 1234 \
   --gradient_checkpointing \
   --zero_stage $ZERO_STAGE \
   --deepspeed \ 

Open-domain QA

/app/anaconda3/bin/deepspeed --num_gpus=1 --master_addr="127.0.0.0" --master_port=29560 Tok-RAG_for_ODQA.py \
   --file_path NQ \ # Your dataset path, such as NQ
   --model_name_or_path $Your_LLM_path \ # Your LLm path
   --seed 1234 \
   --gradient_checkpointing \
   --zero_stage $ZERO_STAGE \
   --deepspeed \