lechmazur/confabulations

Hallucinations (Confabulations) Document-Based Benchmark for RAG. Includes human-verified questions and answers.

29
/ 100
Experimental

This benchmark helps you evaluate how well different large language models (LLMs) avoid making up answers when asked misleading questions based on specific documents. It takes a list of LLMs and a set of challenging questions, then shows you which models are most likely to 'confabulate' or hallucinate, and which tend to simply not respond. An AI developer or researcher building RAG systems would use this to pick the best LLM for their application.

243 stars. No commits in the last 6 months.

Use this if you need to choose an LLM for a Retrieval-Augmented Generation (RAG) system and want to minimize incorrect, made-up answers.

Not ideal if you are looking for an LLM benchmark focused on general accuracy or creative writing rather than specifically confabulation rates.

AI model evaluation RAG system development LLM selection hallucination testing natural language processing
No License Stale 6m No Package No Dependents
Maintenance 2 / 25
Adoption 10 / 25
Maturity 8 / 25
Community 9 / 25

How are scores calculated?

Stars

243

Forks

9

Language

HTML

License

Last pushed

Aug 07, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/rag/lechmazur/confabulations"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.