Multi-Turn Conversation RAG

- Conversation compression method for multi-turn RAG

• Last Response Strategy
• Rewrite Strategy
• LLM Summarization Strategy
• Refert to CORAL Paper

Docker Environment

- Docker Build

docker build -t multi_turn_test_env .

- Docker Run

docker run -it --gpus all --name multi_turn_test_env --shm-size=64G -p {port}:{port} -e GRANT_SUDO=yes --user root -v {root_folder}:/workspace/multi_turn_conversation -w /workspace/multi_turn_conversation multi_turn_test_env bash

- Docker Exec

docker exec -it multi_turn_test_env bash

Multi-turn Conversation RAG Test

- LLM Summarization Strategy

• When using the GPT API, enter the API KEY in the '.env' file created after executing the initialization command

• Last Response Strategy

  • Examples of conversation history file
  • Examples of demonstration file
  • Demonstartion file is not required

python multi_turn_conversation_test_summarization.py --config_path {config_path} --conversation_history_file {conversation_history_file_path} --demonstration_file {demonstration_file_path}

- Simple Rewriting Strategy

• When using the GPT API, enter the API KEY in the '.env' file created after executing the initialization command

• Step 1. Extracting keywords

• Step 2(3). Combine keywords and generate final questions

python multi_turn_conversation_test_simple_rewriting.py --config_path {config_path}

Author

- LinkedIn: https://www.linkedin.com/in/taeyong-kong-016bb2154

- Blog URL: https://blog.naver.com/qbxlvnf11

- Email: qbxlvnf11@google.com, qbxlvnf11@naver.com