Skip to content

Latest commit

 

History

History
119 lines (79 loc) · 4.61 KB

README.md

File metadata and controls

119 lines (79 loc) · 4.61 KB
Raw

🎲 BoardRAG

A RAG application feeding on board games running locally. Create a database with your favorite board game rulesets and query the RAG model to get answers to your questions!

Based on the RAG tutorial by pixegami.

Dependencies

This project is based on the following technologies:

  • Ollama: all the LLMs are run using the Ollama library, as they are run locally.
  • ChromaDB: the database used to store the chunks of the board game rulesets, alongside their corresponding embeddings.
  • Langchain: used for aligning the LLMs calls, the database and the prompt templates.
  • Pytest: for testing the codebase.
  • Pypdf2: for extracting the text from the PDFs.
  • Argilla: for data visualization.

To install the dependencies, please run the following command to create a virtual environment and install the dependencies (instructions for Unix-based systems):

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Ollama requires a separate installation in your computer. Please, go to the Ollama website and follow the instructions to install it.

Dev Dependencies

If you want to contribute to the project, please install the ruff pre-commit hooks by running the following command:

pre-commit install

This will install the pre-commit hooks that will run the tests and the linters before each commit.

Running the application

This RAG application is composed by a process of database population and the RAG model itself. This project also comes along with a testing unit. For executing any of these process, please make sure Ollama is correctly installed in your computer, and the selected LLMs are downloaded.

Vector Database

The first step is to populate the database with the chunks of the board game rulesets and their embeddings. Once you have your desired rulesets in the data folder, please run the following command:

python populate_database.py

This script will extract the text from the PDFs, chunk it, and populate the database with the chunks and their embeddings, calculated by the LLM of your choice (environment variable EMBEDDER_MODEL). If the chunks are not present in the database, they will be added.

If you want to reset the database before updating it, please run the following command:

python populate_database.py --reset

To visualize the chunks generated, you can create an Argilla instance at Hugging Face Spaces and run the following command to populate a dataset with the chunks:

python visualize_db_argilla.py

Make sure that your Hugging Face Spaces API key and URL is stored in the .env file.

Retriever & Generator

The RAG model is composed by a retriever and a generator. The retriever is responsible for finding the most relevant chunks in the database, while the generator is responsible for generating the answer to the user's question.

Once the database is populated, you can run the RAG model by running the following command:

python query.py --query_text "How can I build a hotel in Monopoly?"

You can also include the flags --include_sources and include_contexts to include the sources and chunks used to build the answer, respectively. Remember that the LLM of choice (set by the environment variable GENERATOR_MODEL) must be downloaded in Ollama.

Gradio Interface

If you want to use a Gradio interface to query the RAG model, please run the following command:

python app.py

This will start a Gradio interface where you can input your question and get the answer from the RAG model. Remember that for the Gradio interface to work, the LLM of choice (set by the environment variable GENERATOR_MODEL) must be downloaded in Ollama, and Ollama must be running. The database must also be populated.

Tests

To run the tests, please run the following command:

pytest .

In the test folder, there is a file for each ruleset in the data folder. Remember that, for the tests to run, the GENERATOR_MODEL must be downloaded in Ollama.

Example of .env file

ARGILLA_API_KEY = "YOUR_API_KEY"
ARGILLA_API_URL = "YOUR_API_URL"
CHUNK_OVERLAP = 80
CHUNK_SIZE = 800
CHROMA_PATH = "chroma"
DATA_PATH = "data"
GENERATOR_MODEL = "mistral"
EMBEDDER_MODEL = "nomic-embed-text"
EVAL_TEMPLATE_PATH = "eval_prompt_tests.txt"
JINJA_TEMPLATE_PATH = "rag_query_pixegami.txt"

Future work

  • Visualize chat results using Argilla
  • Build a Gradio interface for the RAG model
  • Dockerize the application
  • Add more rulesets to the database, alongside their test files
  • Include a CI/CD pipeline