Stay on topic with Classifier-Free Guidance
This work addresses the problem of enhancing language model performance and prompt-adherence for users in NLP applications, though it is incremental as it extends an existing technique to new domains.
The paper demonstrates that Classifier-Free Guidance (CFG) can be applied as an inference-time technique in pure language modeling, improving performance across tasks such as Q&A, reasoning, code generation, and machine translation, achieving state-of-the-art results on LAMBADA with LLaMA-7B over PaLM-540B and showing a 75% human preference for GPT4All using CFG over baseline.
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.