Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

LM-Steer

LM-Steer is a method for controlling the style, tone, and behavioral attributes of language model outputs by applying lightweight linear transformations to word embeddings rather than retraining the underlying model. By adjusting the mathematical representations used during token prediction, this approach alters generated text properties such as sentiment, formality, or toxicity while preserving general fluency and coherence. Because it requires training only a minimal fraction of additional parameters relative to the base model size, LM-Steer provides a parameter-efficient alternative to traditional fine-tuning. Furthermore, these learned steering transformations can be scaled continuously to modulate the intensity of a target attribute, combined to steer multiple behavioral traits simultaneously, and transferred across different model architectures.

1 item

Word Embeddings Are Steers for Language Models

Word Embeddings Are Steers for Language Models

Chi Han, Jialiang Xu, Manling Li, Yi Fung, Chenkai Sun, Nan Jiang, Tarek F. Abdelzaher, Heng Ji

Why you should read this

Proposes LM-Steer, a lightweight approach that applies linear transformations to output word embeddings to achieve parameter-efficient, transferable, and continuous style control across language models without retraining internal weights.

Recent studies have shown that large language models (LLMs), when augmented with additional weights through adapters—for instance, LoRA—can learn specialized tasks without updating model parameters—a paradigm known as either Instruction Tuning (IT) for individual instructions or Task Arithmetic (TA) for task-specific adaptation. In such scenarios, different instruction/task datasets induce distinct steering vectors within LLMs' embedding space, shaping behavior according to downstream objectives. Inspired by recent advances in model merging or parameter composition, we investigate whether multiple steering directions (each optimized toward generating responses on one dataset while discouraging outputs outside its distribution via unlearning-like objectives) can be combined into a single vector enabling zero-shot generalization across diverse tasks.

Added

2026-10-02