AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

DRIFT: Decoupled Rollouts and Importance-Weighted Fine-Tuning for Efficient Multi-Turn Optimization

Jian Mu, Tianyi Lin, Chengwei Qin, Zhongxiang Dai, Yao Shu

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 0 citations (OpenAlex)

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
replace online RL rollouts (UFO-5turn) with offline weighted SFT (DRIFT-5turn)Qwen2.5-3B-Instruct · Average over 6 benchmarksmulti@5 accuracy (%): 60.2 -> 60.5Improved
replace online RL rollouts (UFO-5turn) with offline weighted SFT (DRIFT-5turn)Llama3.1-8B-Instruct · Average over 6 benchmarksmulti@5 accuracy (%): 56.1 -> 55.6Hurt
replace UFO with DRIFTLlama3.1-8B-Instruct · MATH500multi@5 accuracy (%): 46.4 -> 48.2Improved
replace UFO with DRIFTLlama3.1-8B-Instruct · General averagemulti@5 accuracy (%): 72.3 -> 70.1Hurt
train with 5-turn offline trajectories instead of single-turn SFTSFT · Average over 6 benchmarksmulti@5 accuracy (%): 51.7 -> 54.9Improved
train with 5-turn offline trajectories instead of single-turn SFTSFT · Average over 6 benchmarksmulti@5 accuracy (%): 49.9 -> 50.0No clear effect
multi-turn RL (UFO-5turn) instead of single-turn PPOPPO · Average over 6 benchmarksmulti@5 accuracy (%): 51.2 -> 60.2Improved
use terminal-step-only supervision instead of weighting all turns (ablation)DRIFT · MATH500accuracyImproved
use moderate beta (0.1) rather than larger or smaller (ablation)DRIFT · cumulative accuracy per turnaccuracyImproved
use smaller discount gamma (ablation)DRIFT · solved rate per turnsolved rateImproved
vary feedback string (e.g. 'Incorrect' vs 'Incorrect. Please think again.') (ablation)DRIFT · cumulative accuracy per turnaccuracyNo clear effect
DRIFT instead of UFO for training cost as turns growDRIFT-5turn · GPU timeGPU timeImproved

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