| dataset_info: | |
| splits: | |
| - name: train | |
| num_examples: 42 | |
| configs: | |
| - config_name: default | |
| data_files: | |
| - split: train | |
| path: endev_testset.jsonl | |
| task_categories: | |
| - question-answering | |
| language: | |
| - en | |
| tags: | |
| - RAG | |
| - RAGAS | |
| - evaluation | |
| - ChaBo | |
| - EnDev | |
| # EnDev RAGAS Test Set | |
| Synthetic Q&A test set (42 pairs) generated with [RAGAS](https://docs.ragas.io) | |
| (`TestsetGenerator.generate_with_chunks`) over chunks sampled from the EnDev corpus | |
| stored in Qdrant collection `endev-bgem3-512` (Gradio-gateway Space | |
| `GIZ/EnDev_Qdrant`). | |
| - Generated: 2026-09-14 UTC | |
| - Generator/judge LLM: Qwen/Qwen3-235B-A22B-Instruct-2507 | |
| - Embeddings: BGE-M3 via the EnDev TEI Inference Endpoint | |
| - Columns: `user_input`, `reference`, `reference_contexts`, `synthesizer_name` | |
| Used to evaluate the deployed ChaBo pipeline at https://giz-endev-orchestrator.hf.space. | |
| Files: `endev_testset.jsonl` (RAGAS-ready), `endev_testset.csv` (flat). | |