AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

ETS: Energy-Guided Test-Time Scaling for Training-Free RL Alignment

Xiuyu Li, Jinkai Zhang, Mingyang Yi, Yu Li, Longqiang Wang, Yue Wang, Ju Fan

ICML 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 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
use ETS energy-guided sampling instead of Best-of-NQwen3-8B · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 66.31 -> 69.34Improved
use ETS instead of Best-of-NQwen3-1.7B · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 51.70 -> 54.76Improved
use ETS instead of Best-of-NLLaDA-8B-Instruct · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 48.90 -> 51.01Improved
replace GRPO post-training with training-free ETSQwen3-8B · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 65.98 -> 69.34Improved
replace GRPO with training-free ETSQwen3-1.7B · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 48.08 -> 54.76Improved
replace VRPO with training-free ETSLLaDA-8B-Instruct · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 46.40 -> 51.01Improved
use importance-sampling acceleration (ETS-IS) instead of full ETSQwen3-8B · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 69.34 -> 66.45Hurt
use ETS-IS instead of full ETSLLaDA-8B-Instruct · Avg (MATH500, GSM8K, GPQA, HumanEval)accuracy: 51.01 -> 49.36Hurt
use ETS instead of Power Sampling on MATH500Qwen3-8B · MATH500accuracy: 70.8 -> 72.4Improved
use ETS instead of Base decodingQwen3-8B · MATH500accuracy: 60.0 -> 72.4Improved
increase total samples M x K (ablation)ETS on Qwen3-8B · HumanEvalaccuracyImproved
increase guidance steps I (ablation)ETS on Qwen3-8B · HumanEvalaccuracyImproved

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