Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866908818222350336 |
|---|---|
| author | Qiu, Xin Gan, Yulu Hayes, Conor F. Liang, Qiyao Xu, Yinggan Dailey, Roberto Meyerson, Elliot Hodjat, Babak Miikkulainen, Risto |
| author_facet | Qiu, Xin Gan, Yulu Hayes, Conor F. Liang, Qiyao Xu, Yinggan Dailey, Roberto Meyerson, Elliot Hodjat, Babak Miikkulainen, Risto |
| contents | Fine-tuning large language models (LLMs) for downstream tasks is an essential stage of modern AI deployment. Reinforcement learning (RL) has emerged as the dominant fine-tuning paradigm, underpinning many state-of-the-art LLMs. In contrast, evolution strategies (ES) has largely been overlooked due to the widespread belief that it does not scale to modern model sizes. This paper overturns this assumption by demonstrating the first successful application of ES to full-parameter fine-tuning of LLMs at the billion-parameter scale, without dimensionality reduction. ES can indeed search over extremely high-dimensional parameter spaces and outperform established RL implementations across multiple axes, including improved tolerance to long-horizon and delayed rewards, robustness across diverse base LLMs, reduced susceptibility to reward hacking, and improved training stability. These findings suggest that ES is not merely a viable alternative to RL, but a fundamentally different and powerful backpropagation-free post-training paradigm that opens a new direction for LLM fine-tuning beyond current RL-based approaches. The source codes are provided at: https://github.com/VsonicV/es-fine-tuning-paper. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_24372 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning Qiu, Xin Gan, Yulu Hayes, Conor F. Liang, Qiyao Xu, Yinggan Dailey, Roberto Meyerson, Elliot Hodjat, Babak Miikkulainen, Risto Machine Learning Artificial Intelligence Neural and Evolutionary Computing Fine-tuning large language models (LLMs) for downstream tasks is an essential stage of modern AI deployment. Reinforcement learning (RL) has emerged as the dominant fine-tuning paradigm, underpinning many state-of-the-art LLMs. In contrast, evolution strategies (ES) has largely been overlooked due to the widespread belief that it does not scale to modern model sizes. This paper overturns this assumption by demonstrating the first successful application of ES to full-parameter fine-tuning of LLMs at the billion-parameter scale, without dimensionality reduction. ES can indeed search over extremely high-dimensional parameter spaces and outperform established RL implementations across multiple axes, including improved tolerance to long-horizon and delayed rewards, robustness across diverse base LLMs, reduced susceptibility to reward hacking, and improved training stability. These findings suggest that ES is not merely a viable alternative to RL, but a fundamentally different and powerful backpropagation-free post-training paradigm that opens a new direction for LLM fine-tuning beyond current RL-based approaches. The source codes are provided at: https://github.com/VsonicV/es-fine-tuning-paper. |
| title | Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning |
| topic | Machine Learning Artificial Intelligence Neural and Evolutionary Computing |
| url | https://arxiv.org/abs/2509.24372 |