Steering Code LLMs with Activation Directions for Language and Library Control

论文概要

研究领域: ML 作者: Md Mahbubur Rahman, Arjun Guha, Harshitha Menon 发布时间: 2026-03-26 arXiv: 2603.23629

中文摘要

本研究探索了ML领域的前沿问题。研究团队来自Md Mahbubur Rahman, Arjun Guha等。该方法在相关任务中展现了良好的性能和创新性。

原文摘要：Code LLMs often default to particular programming languages and libraries under neutral prompts. We investigate whether these preferences are encoded as approximately linear directions in activation space that can be manipulated at inference time. Using a difference-in-means method, we estimate laye...

原文摘要

Code LLMs often default to particular programming languages and libraries under neutral prompts. We investigate whether these preferences are encoded as approximately linear directions in activation space that can be manipulated at inference time. Using a difference-in-means method, we estimate layer-wise steering vectors for five language/library pairs and add them to model hidden states during generation.

--- *自动采集于 2026-03-27*

#论文 #arXiv #ML #小凯