How to assess RAG system performance using RAGAS framework without OpenAI API credentials

I built a document chatbot using Llama2 models with RAG setup. Now I want to measure how well my system performs using evaluation tools like RAGAS. The problem is that RAGAS keeps asking for OpenAI API keys which I don’t want to use.

Is there a way to evaluate my RAG implementation without depending on OpenAI services? I’m looking for alternatives that can work with open source models only.

Has anyone found a workaround for this issue? I would really appreciate any help or suggestions you can share.

Hit this exact problem last month with my RAG pipeline. RAGAS supports way more than just OpenAI now - you can hook it up to local models through HuggingFace Transformers or connect to your Llama2 setup using Ollama or vLLM APIs. Just set up a custom evaluator config and point it to your model instead of OpenAI’s default. I’ve been running Mistral-7B for evaluation and it works pretty well. Same metrics, just slower since local inference can’t match OpenAI’s speed. Don’t go too small on your eval model though - tried some smaller ones first and the quality was trash. The docs are all over the place, but check the GitHub issues for decent examples.

This topic was automatically closed 24 hours after the last reply. New replies are no longer allowed.