Benchmarking

[markNZed]

Hi, the llm-graph-builder is using a sophisticated pipeline and I am wondering if any benchmarking was done e.g. to measure the value of added complexity, tune hyperparameters, and compare with other RAG approaches ? Thanks!

[jexp]

We started to run eval with RAGAS for a set of documents and the different retrievers (but not hyperparameter sets) and also expose a first set of metrics in the UI in the next release.
But not comprehensively so, that's still on todo.

[markNZed]

Thanks for the pointer to RAGAS. I recently got interested in benchmarking RAG and came across:

• https://huggingface.co/datasets/nvidia/ChatRAG-Bench
• https://chatqa-project.github.io/
• https://github.com/naver/bergen

On less related point, it would be nice to have an easy dev setup where changes to frontend code get reflected immediately and restarting the back end is automatic or simple. Maybe a devcontainer - the ability to run something in a codespace through Git makes projects very accessible.

[kartikpersistent]

we have implemented rags evaluation in our application