Matching Accuracy, Different Geometry: Evolution Strategies vs GRPO in LLM Post-Training

• URL: http://arxiv.org/abs/2604.01499v1
• Date: Thu, 02 Apr 2026 00:16:48 GMT
• Title: Matching Accuracy, Different Geometry: Evolution Strategies vs GRPO in LLM Post-Training
• Abstract summary: Evolution Strategies (ES) have emerged as a scalable gradient-free alternative to reinforcement learning based fine-tuning.<n>We compare ES and Group Relative Policy Optimization across four tasks in both single-task and sequential continual-learning settings.<n>ES matches or exceeds GRPO in single-task accuracy and remains competitive sequentially when its budget is controlled.
• Score: 9.692209933810185
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Evolution Strategies (ES) have emerged as a scalable gradient-free alternative to reinforcement learning based LLM fine-tuning, but it remains unclear whether comparable task performance implies comparable solutions in parameter space. We compare ES and Group Relative Policy Optimization (GRPO) across four tasks in both single-task and sequential continual-learning settings. ES matches or exceeds GRPO in single-task accuracy and remains competitive sequentially when its iteration budget is controlled. Despite this similarity in task performance, the two methods produce markedly different model updates: ES makes much larger changes and induces broader off-task KL drift, whereas GRPO makes smaller, more localized updates. Strikingly, the ES and GRPO solutions are linearly connected with no loss barrier, even though their update directions are nearly orthogonal. We develop an analytical theory of ES that explains all these phenomena within a unified framework, showing how ES can accumulate large off-task movement on weakly informative directions while still making enough progress on the task to match gradient-based RL in downstream accuracy. These results show that gradient-free and gradient-based fine-tuning can reach similarly accurate yet geometrically distinct solutions, with important consequences for forgetting and knowledge preservation. The source code is publicly available: https://github.com/Bhoy1/ESvsGRPO.

Related papers

• Understanding Evolution Strategies for LLM Reasoning: Broader Reasoning Coverage than GRPO [29.055105879127904]
  Evolution Strategies (ES) have emerged as a memory-efficient post-training paradigm for reasoning.<n>This paper first identifies a performance advantage of ES over Group Relative Policy Optimization ( GRPO)<n>We develop a sequential GRPO-ES training strategy that combines GRPO's strength in Pass@1 with ES's gains in Pass@K.
  arXiv  Detail & Related papers  (2026-08-27T16:48:39Z)
• Group-Graph Policy Optimization for Long-Horizon Agentic Reinforcement Learning [85.02566758103008]
  Group-Graph Policy Optimization (G2PO) is a novel group-based Reinforcement Learning algorithm tailored for multi-turn agentic tasks.<n>G2PO substantially outperforms state-of-the-art prompt-based and RL baselines, achieving remarkable success rate improvements of up to 22.2% over GRPO.
  arXiv  Detail & Related papers  (2026-06-22T08:12:47Z)
• FBOS-RL: Feedback-Driven Bi-Objective Synergistic Reinforcement Learning [16.200486964371713]
  We propose FBOS-RL, a Feedback-Driven Bi-Objective Synergistic reinforcement learning framework.<n>Specifically, we let the model perform Feedback-Guided Exploration Enhancement based on the feedback provided by the environment.<n>Under an identical number of rollouts, FBOS-RL learns substantially faster than GRPO and feedback-based baselines.
  arXiv  Detail & Related papers  (2026-05-18T12:48:36Z)
• TopoCurate:Modeling Interaction Topology for Tool-Use Agent Training [53.93696896939915]
  Training tool-use agents typically rely on Supervised Fine-Tuning (SFT) on successful trajectories and Reinforcement Learning (RL) on pass-rate-selected tasks.<n>We propose TopoCurate, an interaction-aware framework that projects multi-trial rollouts from the same task into a unified semantic quotient topology.<n>TopoCurate achieves consistent gains of 4.2% (SFT) and 6.9% (RL) over state-of-the-art baselines.
  arXiv  Detail & Related papers  (2026-03-02T10:38:54Z)
• Evolutionary Strategies lead to Catastrophic Forgetting in LLMs [51.91763220981834]
  Evolutionary Strategies (ES) have recently re-emerged as a gradient-free alternative to traditional learning algorithms.<n>ES is able to reach performance numbers close to GRPO for math and reasoning tasks with a comparable compute budget.<n>ES is accompanied by significant forgetting of prior abilities, limiting its applicability for training models online.
  arXiv  Detail & Related papers  (2026-01-28T18:59:34Z)
• TL-GRPO: Turn-Level RL for Reasoning-Guided Iterative Optimization [97.18886232580131]
  Large language models have demonstrated strong reasoning capabilities in complex tasks through tool integration.<n>We propose Turn-Level GRPO, a lightweight RL algorithm that performs turn-level group sampling for fine-grained optimization.
  arXiv  Detail & Related papers  (2026-01-23T06:21:33Z)
• Inpainting-Guided Policy Optimization for Diffusion Large Language Models [67.97530437998117]
  Masked diffusion large language models (dLLMs) are emerging as promising alternatives to autoregressive LLMs.<n>We explore how inpainting can inform RL algorithm design for dLLMs.
  arXiv  Detail & Related papers  (2025-09-12T16:44:31Z)
• TGRPO :Fine-tuning Vision-Language-Action Model via Trajectory-wise Group Relative Policy Optimization [12.061547251822326]
  Trajectory-based Group Relative Policy Optimization (TGRPO) is an online RL-based training framework for Visual-Language-Action (VLA) models.<n>We show that TGRPO achieves an average success rate of 80.7%, which is 4.2% higher than that of Supervised Fine-Tuning (SFT) and outperforms other representative RL-based post-training methods.
  arXiv  Detail & Related papers  (2025-06-10T04:27:49Z)
• Stepwise Guided Policy Optimization: Coloring your Incorrect Reasoning in GRPO [22.00487909203855]
  Group Relative Policy Optimization fails to update a policy when all responses within a group are incorrect.<n>This limitation underscores a key gap between artificial and human intelligence.<n>We introduce a simple framework that mitigates the all-negative-sample issue by incorporating response diversity within groups.
  arXiv  Detail & Related papers  (2025-05-16T18:02:05Z)
• Improving Large Language Model Planning with Action Sequence Similarity [50.52049888490524]
  In this work, we explore how to improve the model planning capability through in-context learning (ICL)<n>We propose GRASE-DC: a two-stage pipeline that first re-samples high AS exemplars and then curates the selected exemplars.<n>Our experimental result confirms that GRASE-DC achieves significant performance improvement on various planning tasks.
  arXiv  Detail & Related papers  (2025-05-02T05:16:17Z)
• LiNeS: Post-training Layer Scaling Prevents Forgetting and Enhances Model Merging [80.17238673443127]
  LiNeS is a post-training editing technique designed to preserve pre-trained generalization while enhancing fine-tuned task performance.<n>LiNeS demonstrates significant improvements in both single-task and multi-task settings across various benchmarks in vision and natural language processing.
  arXiv  Detail & Related papers  (2024-10-22T16:26:05Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.