首页 > AI前沿 > Get Hint-Augmented GRPO Done Right Through Activation Steering

Get Hint-Augmented GRPO Done Right Through Activation Steering

arXiv机器学习 2026-11-18 01:43 7 阅读 查看原文

背景与问题

GRPO has become the dominant reinforcement learning method for eliciting reasoning capability in LLMs using verifiable rewards. However, it suffers from a fundamental failure mode: when a training question exceeds the base model's current capability, all generated responses fail and receive zero reward, yielding no gradient signal and stalling further capability expansion.

HA-GRPO的局限

An emerging class of GRPO extensions, termed Hint-Augmented GRPO (HA-GRPO), adds a "hint" to the context to reduce the difficulty of the question, thereby producing a usable gradient signal. Nevertheless, existing HA-GRPO methods share a key issue: no hint is available at inference time to append to the question, creating a training-inference context mismatch that leads to inconsistent performance gains.

ASH方法

In this work, we reinterpret the function of hinting -- not as reducing question difficulty, but as steering model behavior -- and propose ASH (Activation Steering for Hinting), a novel HA-GRPO method built on this view. During training, the model is presented with both hinted and plain questions, while a trainable steering vector is optimized to widen the margins between the model's generation probabilities for responses to the hinted version relative to the plain version, thereby capturing the question-agnostic component of the behavioral shift induced by the hint. During inference, the trained steering vector is directly injected into the model's activation, reproducing the hint-conditioned behavior without requiring any hint in the context.

实验结果

Experiments with four LLMs (Qwen3-1.7B/4B-Base, Llama-3.2-1B/3B-Instruct) show that ASH consistently improves pass@$k$ on competition-level mathematics benchmarks (AMC, OlympiadBench, AIME24, AIME25) and out-of-distribution benchmarks (Minerva, GPQA-Diamond). Notably, it improves average pass@8 of Qwen3-1.7B-Base and Qwen3-4B-Base by 2.60% and 2.21%.