Research map / Large language models / Retrieval and knowledge
RAG-R1:Incentivizing the Search and Reasoning Capabilities of LLMs Through Multi-Query Parallelism
Zhiwen Tan, Jiaming Huang, Qintong Wu, Hongxuan Zhang, Chenyi Zhuang, 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 |
|---|---|---|---|
| add format-learning SFT before RL (ablation) | RAG-R1-sq · multi-hop QA average | EM: 0.318 -> 0.410 | Improved |
| add RL stage after SFT (ablation) | RAG-R1-sq · multi-hop QA average | EM: 0.344 -> 0.410 | Improved |
| add RL data selection (filter unanswerable samples) (ablation) | RAG-R1-sq · multi-hop QA average | EM: 0.370 -> 0.410 | Improved |
| add RL data selection (ablation) | RAG-R1-mq · multi-hop QA average | EM: 0.411 -> 0.423 | Improved |
| add format-learning SFT before RL (ablation) | RAG-R1-mq · multi-hop QA average | EM: 0.339 -> 0.423 | Improved |
| move from single-query to multi-query parallelism | RAG-R1 · 7 QA benchmarks average | EM: 0.449 -> 0.457 | Improved |
| move from single-query to multi-query parallelism | RAG-R1 · NQ | EM: 0.429 -> 0.423 | Hurt |
| move from single-query to multi-query parallelism | RAG-R1 · 2WikiMultiHopQA | EM: 0.520 -> 0.563 | Improved |
| move from single-query to multi-query parallelism | RAG-R1 · inference time | seconds: 8.21 -> 7.42 | Improved |
| use multi-query instead of single-query in prompted Qwen2.5-72B-Instruct | Search-R1-style agent · 2WikiMultiHopQA | EM (%): 45.59 -> 50.45 | Improved |
| use multi-query instead of single-query in prompted Qwen2.5-72B-Instruct | Search-R1-style agent · HotpotQA | EM (%): 43.79 -> 44.33 | Improved |
| use RAG-R1-mq instead of R1-Searcher | RL-based search agents · 7 QA benchmarks average | EM: 0.402 -> 0.457 | Improved |
Closest papers by meaning
- AceSearcher: Bootstrapping Reasoning and Search for LLMs via Reinforced Self-Play (NeurIPS 2025 · similarity 0.90)
- Improving Retrieval-Augmented Generation through Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Boosting the Potential of Large Language Models with an Intelligent Information Assistant (NeurIPS 2024 · similarity 0.90)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.89)
- Hogwild! Inference: Parallel LLM Generation via Concurrent Attention (NeurIPS 2025 · similarity 0.89)
- ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
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.
