violenil / jina-rag

RAG with Jina Embeddings and LlamaIndex

This repo contains a notebook that guides you through the process of using Jina Embeddings V2 in a retrieval augmented generation pipeline. It uses the model jinaai/jina-embeddings-v2-base-de from HuggingFace, which is a bilingual German-English embedding model trained by Jina AI.

Install the requirements in requirements.txt in a fresh environment.

To run the notebook rag.ipynb, it is also necessary to host a service that serves a (small) large language model. You can do this on an ordinary local machine without GPUs using ollama. To setup such a service:

1. Follow the installation instruction for ollama
2. Pull the mistral (or any other small) LLM model using ollama pull mistral
3. Start a service by running ollama serve

Now you can execute all blocks in the notebook.

Disclaimer: The data in the data/ folder was generated using GPT4 for the purpose of demonstrating the RAG pipeline on unseen, "private" data.