Stay on topic with Classifier-Free Guidance

TL;DR

This paper extends Classifier-Free Guidance from text-to-image to pure language modeling, demonstrating its ability to enhance performance, improve coherence, and complement other inference techniques across various NLP tasks.

Contribution

It introduces the novel application of CFG as an inference-time method in language models, showing significant performance gains and compatibility with other techniques.

Findings

CFG improves performance across multiple language models and tasks.

CFG achieves state-of-the-art results on LAMBADA with smaller models.

CFG enhances faithfulness and coherence in human evaluations.

Abstract

Classifier-Free Guidance (CFG) has recently emerged in text-to-image generation as a lightweight technique to encourage prompt-adherence in generations. In this work, we demonstrate that CFG can be used broadly as an inference-time technique in pure language modeling. We show that CFG (1) improves the performance of Pythia, GPT-2 and LLaMA-family models across an array of tasks: Q\&A, reasoning, code generation, and machine translation, achieving SOTA on LAMBADA with LLaMA-7B over PaLM-540B; (2) brings improvements equivalent to a model with twice the parameter-count; (3) can stack alongside other inference-time methods like Chain-of-Thought and Self-Consistency, yielding further improvements in difficult tasks; (4) can be used to increase the faithfulness and coherence of assistants in challenging form-driven and content-driven prompts: in a human evaluation we show a 75\% preference for GPT4All using CFG over baseline.