3,457 2 months ago

A local evaluator for retrieval-augmented generation (RAG) answers.

tools 1b 3b
ollama run erukude/rag-audit-judge:3b

Details

2 months ago

363fe372386d · 2.1GB ·

llama
·
3.21B
·
Q4_K_M
You are RAG-AUDIT-JUDGE, a strict evaluator for retrieval-augmented generation answers. Return exact
{ "num_ctx": 4096, "stop": [ "<|start_header_id|>", "<|end_header_id|>",
llama
·
48.6M
LLAMA 3.2 COMMUNITY LICENSE AGREEMENT Llama 3.2 Version Release Date: September 25, 2024 “Agreemen
**Llama 3.2** **Acceptable Use Policy** Meta is committed to promoting safe and fair use of its tool
<|start_header_id|>system<|end_header_id|> Cutting Knowledge Date: December 2023 {{ if .System }}{{

Readme

RAG Audit Judge

Base model: Llama3.2.

It checks:

  • Is the answer grounded in the supplied context?
  • Is reasoning unsupported?
  • Are citations missing or invalid?
  • Is the response overconfident?
  • Is sensitive information exposed?

It returns one JSON object:

{
  "groundedness_score": 0.82,
  "hallucination_risk": "low",
  "citation_quality": "medium",
  "issues": [],
  "recommended_fix": ""
}

Provide a question, retrieved context entries with stable source IDs, and the candidate answer. The judge evaluates only the supplied context, not outside knowledge.

Example user input:

Question: Can I return a headset?

Context:
- id: refund-policy
  text: Headsets can be returned within 30 days if they are unopened.

Candidate answer:
An unopened headset can be returned within 30 days, and a $5 restocking fee always applies [refund-policy].

Author

Author: Sai Teja Erukude
Role: Developer & Maintainer