Experiment with Langchain to build converational RAG Q&A application.

The application can use any LLM that can be hosted by OpenAI compatible API server.

1. Edit config.yaml by set openai_api_key to OpenAI API key, and remove value openai_compatible_base_url (make it empty value).
2. From root directory of the project, run docker compose -f docker/docker-compose.yaml up -d.
3. Access the web via port 8501 (you can edit the port mapping in docker/docker-compose.yaml).

1. Download .gguf model from Hugging Face. I use bartowski/Meta-Llama-3.1-8B-Instruct-GGUF Q6_K_L model, and put it in models directory.
2. Edit docker/docker-compose.llama-cpp.yaml file to config how llama.cpp server run as follow.
   • In image attribute, set it to image version you want to use. I build it locally, so I use local/llama.cpp:server-cuda. Generally, the server image is enough for running e.g. ghcr.io/ggerganov/llama.cpp:server
   • In volumes attribute, add bind mount as necessary to add document to use for RAG.
   • In command attribute, change Meta-Llama-3.1-8B-Instruct-Q6_K_L.gguf to the model name that you download into models directory.
   • In command attribute, change any necessary you want to config the server.
3. In docker/docker-compose.yaml file, edit volumes attribute to add bind mount as necessary to add document to use for RAG.
4. Edit config.yaml file to config how the app run as follow
   • Set openai_api_key to any string (but not empty string).
   • Set openai_compatible_base_url to http://llama-cpp-server:8000.
   • Set document_path to the path of the document that you add to bind mount.
5. From root directory of the project, run docker compose -f docker/docker-compose.yaml -f docker/docker-compose.llama-cpp.yaml up -d.