FreeCtrl: Constructing Control Centers with Feedforward Layers for Learning-Free Controllable Text Generation

TL;DR

FreeCtrl is a novel learning-free method that adjusts feedforward neural network weights to control large language model outputs, achieving effective attribute control without extensive training or data.

Contribution

It introduces a dynamic, learning-free approach to controllable text generation by manipulating FFN weights, outperforming existing learning-free and learning-based methods.

Findings

FreeCtrl effectively controls attribute keywords in generated text.

It outperforms other learning-free methods in attribute control tasks.

The approach reduces computational costs compared to traditional learning-based methods.

Abstract

Controllable text generation (CTG) seeks to craft texts adhering to specific attributes, traditionally employing learning-based techniques such as training, fine-tuning, or prefix-tuning with attribute-specific datasets. These approaches, while effective, demand extensive computational and data resources. In contrast, some proposed learning-free alternatives circumvent learning but often yield inferior results, exemplifying the fundamental machine learning trade-off between computational expense and model efficacy. To overcome these limitations, we propose FreeCtrl, a learning-free approach that dynamically adjusts the weights of selected feedforward neural network (FFN) vectors to steer the outputs of large language models (LLMs). FreeCtrl hinges on the principle that the weights of different FFN vectors influence the likelihood of different tokens appearing in the output. By identifying and adaptively adjusting the weights of attribute-related FFN vectors, FreeCtrl can control the output likelihood of attribute keywords in the generated content. Extensive experiments on single- and multi-attribute control reveal that the learning-free FreeCtrl outperforms other learning-free and learning-based methods, successfully resolving the dilemma between learning costs and model performance.