AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

FrugalRAG: Less is More in RL Finetuning for Multi-hop Question Answering

Abhinav Java, Srivathsan Koundinyan, Nagarajan Natarajan, Amit Sharma

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Explanatory and multiple-choice QA benchmarks

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 changedWhereMeasuredEffect
add RL stage 2 (learn to stop) after exploration SFT (ablation)FrugalRAG-Explore · HotPotQA (ColBERTv2)searches per question: 5.99 -> 2.05Improved
add RL stage 2 after exploration SFT (ablation)FrugalRAG-Explore · MuSiQue (ColBERTv2)MBE accuracy: 30.10 -> 33.72Improved
add RL stage 2 after exploration SFT (ablation)FrugalRAG-Explore · HotPotQA (ColBERTv2)recall: 86.40 -> 82.80Hurt
replace SFT-with-finish baseline by RL stage 2 (ablation)SFT (with finish) · 2Wiki (E5-base-v2)efficiency tradeoff: 13.85 -> 18.02Improved
replace SFT-with-finish by RL stage 2 (ablation)SFT (with finish) · HotPotQA (E5-base-v2)efficiency tradeoff: 22.31 -> 22.30No clear effect
add stage-1 SFT before RL (ablation)FrugalRAG · HotPotQArecall: 66.5 -> 70.4Improved
add late-stop penalty to reward (ablation)FrugalRAG · HotPotQAavg searches: 3.144 -> 2.89Improved
jointly optimize query generation and stopping with recall reward (ablation)FrugalRAG · HotPotQArecall: 70.4 -> 68.3Hurt
RL directly on base model with recall reward (ablation)FrugalRAG · HotPotQArecall: 70.4 -> 66.8Hurt
swap answer generator Qwen2.5-7B for Qwen2.5-32BFrugalRAG · HotPotQA (E5-base-v2)MBE accuracy: 58.5 -> 61.4Improved
swap Qwen2.5-7B-Instruct base for Llama3.1-8B-Instruct (ablation)FrugalRAG · HotPotQArecall: 70.40 -> 70.00No clear effect
scale training examples from 100 to 500 (ablation)FrugalRAG · HotPotQArecall: 67.9 -> 70.8Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy