AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Getting Your LLMs Ready for Reinforcement Learning with Lightweight SFT

Xinran Li, Guangda Huzhang, Siqi Shen, Qing-Guo Chen, Zhao Xu, Weihua Luo, Kaifu Zhang, Jun Zhang

ICLR 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 AESL cold-start instead of CE cold-startGRPO post-training · Qwen2.5-7B-Instruct, Math avg after RLaverage: 40.19 -> 42.26Improved
use AESL cold-start instead of CE cold-startGRPO post-training · Qwen2.5-Math-7B, Math avg after RLaverage: 48.6 -> 50.04Improved
use AESL cold-start instead of CE cold-startGRPO post-training · Llama-3.1-8B-Instruct, Math avg after RLaverage: 22.46 -> 25.5Improved
early-stop CE cold-start at the diversity turning pointCE cold-start · Qwen2.5-7B-Instruct, Math avg after RLaverage: 40.19 -> 41.54Improved
add cold-start SFT before RL (AESL)GRPO · Qwen2.5-7B-Instruct, Math avg after RLaverage: 38.09 -> 42.26Improved
use prefix-confidence denominator instead of constant denominator (ablation)AESL · Qwen2.5-7B-Instruct, Math avg after RLaverage: 40.35 -> 42.26Improved
use soft weighting instead of masking top-1 tokens (ablation)AESL · Qwen2.5-7B-Instruct, Math avg after RLaverage: 40.6 -> 42.26Improved
use AESL weighting instead of linear inverse weighting (ablation)AESL · Qwen2.5-7B-Instruct, Math avg after RLaverage: 40.96 -> 42.26Improved
increase t_scaling from 1 to 5 (ablation)AESL · Qwen2.5-7B-Instruct, Math avg after RLaverage: 41.09 -> 42.26Improved
increase t_scaling from 5 to 7 (ablation)AESL · Qwen2.5-7B-Instruct, Math avg after RLaverage: 42.26 -> 41.1Hurt
increase cold-start data from 1k to 6k (ablation)AESL · Qwen2.5-7B-Instruct, Math avg after RLaverage: 41.14 -> 42.79Improved
use AESL instead of CE-ES on simple questionscold-start · Qwen2.5-7B-Instruct, simple 3k Math avg after RLaverage: 39.35 -> 42.23Improved

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