Proxy Exploration and Reusable Guidance: A Modular LLM Post-Training Paradigm via Proxy-Guided Update Signals

2026-07-13

AI Summary

This work addresses the high computational cost and limited reusability of optimization signals in existing post-training methods for large language models, which tightly couple policy exploration with distribution alignment. The authors propose PUST, a novel framework that fully decouples these two components for the first time. PUST employs a lightweight proxy model to efficiently explore policy improvements, extracts directional update signals from its relative performance gains before and after optimization, and transfers these signals to the main model. This approach enables tunable enhancement from weak to strong models, supports caching and reuse of optimization signals, and facilitates cross-model transfer. Experiments on Qwen3 variants across mathematical and coding tasks demonstrate that even substantially weaker proxies can consistently boost main model performance while significantly reducing computational overhead, establishing a modular and efficient paradigm for post-training.

Abstract

Post-training is essential for refining the domain-specific capabilities of large language models (LLMs), yet existing reward optimization and distribution matching methods tightly couple policy exploration with distribution alignment. This coupling forces expensive exploration directly on the policy model and severely hinders the asynchronous generation, reuse, and cross-model transfer of optimization signals. In this paper, we propose Proxy-guided Update Signal Transfer (PUST), a novel post-training framework that fundamentally decouples update-signal exploration from distribution alignment. Instead of utilizing the primary model for costly exploration, PUST employs a lightweight proxy model as an efficient testbed to discover high-reward behaviors. We extract the relative improvement signal between the proxy's initial and optimized states, transferring this directional update to the primary model to guide its policy alignment. This decoupled pipeline, comprising proxy exploration, update-signal extraction, and signal transfer, significantly reduces computational overhead and enables optimization signals to be asynchronously generated, cached, and reused. Crucially, by transferring relative improvements rather than absolute policy distributions, PUST naturally supports weak-to-strong improvement and seamless cross-model transfer. Systematic evaluations on Qwen3-family models across math and code domains demonstrate that update signals extracted from substantially weaker proxies can robustly and adjustably enhance stronger primary models. Ultimately, PUST transforms post-training from a monolithic online optimization process into a highly modular, reusable, and cost-efficient paradigm.

Problem

• post-training
• policy exploration
• distribution alignment
• optimization signals
• cross-model transfer

Innovation

• proxy-guided learning
• modular post-training
• update signal transfer
• relative policy improvement
• cross-model transfer