RAGLight

RAGLight is a lightweight and modular Python library for implementing Retrieval-Augmented Generation (RAG). It enhances the capabilities of Large Language Models (LLMs) by combining document retrieval with natural language inference.

Designed for simplicity and flexibility, RAGLight provides modular components to easily integrate various LLMs, embeddings, and vector stores, making it an ideal tool for building context-aware AI solutions. ✨

---

Features 🔥

• 🌐 Embeddings Model Integration: Plug in your preferred embedding models (e.g., HuggingFace all-MiniLM-L6-v2) for compact and efficient vector embeddings.
• 🧙🏽 LLM Agnostic: Seamlessly integrates with different LLMs, such as llama3 or custom providers, for natural language inference.
• ⚖️ RAG Pipeline: Combines document retrieval and language generation in a unified workflow.
• 🖋️ Flexible Document Support: Ingest and index various document types (e.g., PDF, TXT, DOCX).
• 🛠️ Extensible Architecture: Easily swap vector stores, embedding models, or LLMs to suit your needs.

---

Installation 🛠️

Install RAGLight directly from PyPI:

pip install raglight

---

Quick Start 🚀

1. Configure Your Pipeline

Set up the components of your RAG pipeline:

from raglight.rag.builder import Builder
from raglight.config.settings import Settings

rag = Builder() \
    .with_embeddings(Settings.HUGGINGFACE, model_name=model_embeddings) \
    .with_vector_store(Settings.CHROMA, persist_directory=persist_directory, collection_name=collection_name) \
    .with_llm(Settings.OLLAMA, model_name=model_name, system_prompt_file=system_prompt_directory) \
    .build_rag()

2. Ingest Documents

Use the pipeline to ingest documents into the vector store:

rag.vector_store.ingest(file_extension='**/*.pdf', data_path='./data')

3. Query the Pipeline

Retrieve and generate answers using the RAG pipeline:

response = rag.question_graph("How can I optimize my marathon training?")
print(response)

---

Advanced Configuration ⚙️

Environment Variables

Configure the pipeline with environment variables for better modularity:

export PERSIST_DIRECTORY=./vectorstore
export MODEL_EMBEDDINGS=all-MiniLM-L6-v2
export MODEL_NAME=llama3

You can also define these in a .env file:

PERSIST_DIRECTORY=./vectorstore
MODEL_EMBEDDINGS=all-MiniLM-L6-v2
MODEL_NAME=llama3

---

TODO

• Feature: Add the possibility to use custom pipelines while ingesting data into the Vector Store.
• Feature: Add support for new Vector Stores (e.g., FAISS, Weaviate, Milvus).
• Feature: Integrate new LLM providers (e.g., VLLM, HuggingFace, GPT-Neo).

---

🚀 Get started with RAGLight today and build smarter, context-aware AI solutions!