AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning

Xin Qiu, Yulu Gan, Conor F. Hayes, Qiyao Liang, Yinggan XU, Roberto Dailey, Elliot Meyerson, Babak Hodjat, Risto Miikkulainen

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression · 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 ES instead of Dr.GRPO-vQwen-2.5-3B-Instruct · Countdownaccuracy (%): 43.8 -> 60.5Improved
use ES instead of GRPO-vLlama-3.1-8B-Instruct · Countdownaccuracy (%): 46.9 -> 61.2Improved
use ES instead of Dr.GRPO-vQwen-2.5-0.5B-Instruct · Countdownaccuracy (%): 13.5 -> 14.4Improved
use ES instead of GRPO-v over three seedsQwen2.5-3B-Instruct · Countdownaccuracy (%): 35.07 -> 58.67Improved
use ES instead of GRPO with KL penaltyQwen-2.5-7B-Instruct · Concisenessreward and KL: reward 0.911 -> 0.919; KL 1.591 -> 0.813Improved
remove KL penalty from GRPO (beta 0.0 and 0.01) (ablation)GRPO · Concisenessreward hackingHurt
use ES instead of GRPO for run-to-run consistencyQwen-2.5-7B-Instruct · Concisenessstd of reward: 15.5x higher standard deviation for GRPOImproved
ES fine-tune base model on ARC-AGIQwen-2.5-14B · ARC-AGIaccuracy (%): 0.2 -> 29.5Improved
ES fine-tune base model on SudokuQwen-2.5-3B · Sudokuaccuracy (%): 2.5 -> 69.5Improved
increase ES population size from 16 to 128 (ablation)ES · GSM8K Qwen2.5-1.5Bpass@1: 0.697 -> 0.727Improved
increase ES population size from 16 to 128 (ablation)ES · GSM8K Qwen2.5-3Bpass@1: 0.807 -> 0.831Improved
increase ES population size from 30 to 64 (ablation)ES · GSM8K Qwen2.5-3Bpass@1: 0.823 -> 0.804Hurt

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