Research map / Large language models / Retrieval and knowledge
PRGB Benchmark: A Robust Placeholder-Assisted Algorithm for Benchmarking Retrieval-Augmented Generation
Zhehao Tan, Yihan Jiao, Dan Yang, Junwei Liu, Lei Liu, Jie Feng, Duolin Sun, Yue Shen, Jian Wang, Peng Wei, Jinjie Gu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace weak-only noise with harder noise mix (1:3:1) (ablation) | Qwen2.5 7B · PRGB Filtering-zh | accuracy: 95.33 -> 80.83 | Hurt |
| replace weak-only noise with harder noise mix (1:3:1) (ablation) | Qwen2.5 72B · PRGB Filtering-zh | accuracy: 95.58 -> 92.92 | Hurt |
| replace weak-only noise with harder noise mix (1:3:1) (ablation) | Qwen2.5 7B · PRGB Filtering-en | accuracy: 89.01 -> 84.72 | Hurt |
| replace weak-only noise with harder noise mix (1:3:1) (ablation) | Qwen2.5 72B · PRGB Filtering-en | accuracy: 89.09 -> 87.42 | Hurt |
| enable thinking mode | Qwen3-235B-A22B · PRGB ZH Overall | accuracy: 75.20 -> 80.76 | Improved |
| enable thinking mode | Qwen3-30B-A3B · PRGB ZH Overall | accuracy: 71.05 -> 80.45 | Improved |
| enable thinking mode | Qwen3-235B-A22B · PRGB EN Overall | accuracy: 70.27 -> 78.68 | Improved |
| enable thinking mode | Qwen3-8B · PRGB ZH Overall | accuracy: 50.02 -> 63.04 | Improved |
| scale model from 7B to MAX | Qwen2.5 · PRGB Combination L3 | accuracy: 28.60 -> 75.20 | Improved |
| scale model from 7B to MAX | Qwen2.5 · PRGB Combination L1 | accuracy: 74.07 -> 86.53 | Improved |
| scale model from 7B to MAX | Qwen2.5 · PRGB Reference Reasoning Overall | accuracy: 26.92 -> 55.37 | Improved |
| scale model from 32B to 72B | Qwen2.5 · PRGB Reference Reasoning Overall | accuracy: 44.60 -> 44.14 | No clear effect |
Closest papers by meaning
- RAGGED: Towards Informed Design of Scalable and Stable RAG Systems (ICML 2025 · similarity 0.94)
- Benchmarking Large Language Models in Retrieval-Augmented Generation (AAAI 2024 · similarity 0.93)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.93)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (ICLR 2025 · similarity 0.92)
- Inference Scaling for Long-Context Retrieval Augmented Generation (ICLR 2025 · similarity 0.92)
- A Theory for Token-Level Harmonization in Retrieval-Augmented Generation (ICLR 2025 · similarity 0.92)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
