RAG QA Logs & Corpus — Multi-Table Dataset for Retrieval QA Evaluation
#3
by
tarekmasryo - opened
🔎 Production-style synthetic RAG evaluation data: corpus → chunks → retrieval/rerank → answers → judgments.
📦 Dataset snapshot
- Multi-table package with joinable IDs (doc/chunk/query/run-style keys)
- Includes evaluation outcomes (correctness / hallucination / faithfulness-style signals)
🔎 What’s inside
- Corpus + chunked text tables
- Query/log tables (runs, scenarios, strategies)
- Retrieval and ranking outputs (top-k style results)
- Answer + evaluation/judgment tables
✅ Good for
📊 RAG dashboards (quality × latency × cost trade-offs)
🧪 Comparing retrieval strategies (BM25 vs dense vs hybrid; with/without rerank)
🤖 Risk scoring prototypes (predict likely failure/hallucination from telemetry)
⚠️ Notes
- Fully synthetic benchmark → great for workflows and regression testing examples.
- Keep strict group-safe splits when training/validating to avoid contamination.
🧾 Not for high-stakes decisions (clinical/legal/financial).