project-miracl/nomiracl

NoMIRACL: A multilingual hallucination evaluation dataset to evaluate LLM robustness in RAG against first-stage retrieval errors on 18 languages.

38
/ 100
Emerging

The dataset includes both relevant and non-relevant query-passage pairs balanced at 50/50, enabling evaluation of whether LLMs can correctly abstain when no relevant information is retrieved. It provides a unified Python interface supporting multiple inference backends (HuggingFace, vLLM, Cohere, OpenAI, NVIDIA, Azure, Anyscale) and includes four customizable prompt templates (vanilla, role-based, repetition, explanation) for systematic robustness testing across techniques.

No commits in the last 6 months. Available on PyPI.

Stale 6m
Maintenance 0 / 25
Adoption 7 / 25
Maturity 18 / 25
Community 13 / 25

How are scores calculated?

Stars

26

Forks

4

Language

Python

License

Apache-2.0

Last pushed

Nov 29, 2024

Commits (30d)

0

Dependencies

6

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/rag/project-miracl/nomiracl"

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