Consolidating Rewarded Perturbations for LLM Post-Training 文章

ArXiv CS.CL2026-06-01NEWSen作者: Zheyu Zhang, Shuo Yang, Gjergji Kasneci

摘要

arXiv:2605.31494v1 Announce Type: new Abstract: Post-training of language models is commonly framed as a sample-score-update loop implemented by gradient descent. A recent line of work, exemplified by RandOpt, relocates this loop to weight space, sampling Gaussian perturbations around a pretrained model and ensembling the top-K rewarded specialists at inference. While competitive with PPO and GRPO under matched training compute, this prediction-level ensemble incurs K forward passes per test example and does not extend cleanly to free-form generation. We ask whether the rewarded population can instead be folded into a single deployable model, replacing the inference-time ensemble with one consolidated update. A split-half analysis over 25 model-task pairs reveals reproducible low-rank structure in every case.

相关事件查看全部 (1)

相关公司

暂无数据

相关人物

暂无数据

相关产品

暂无数据