GRACE: Discriminator-Guided Chain-of-Thought Reasoning
This addresses the issue of unreliable multi-step reasoning in language models for tasks like math and symbolic reasoning, offering an incremental improvement over existing methods.
The paper tackles the problem of language models assigning high likelihood to incorrect steps in chain-of-thought reasoning, proposing GRACE, a stepwise decoding approach that uses a discriminator to guide reasoning towards correct steps, resulting in substantial performance gains on math and symbolic reasoning tasks, such as outperforming baselines by sizeable margins when combined with self-consistency.
In the context of multi-step reasoning, e.g., with chain-of-thought, language models (LMs) can easily assign a high likelihood to incorrect steps. As a result, decoding strategies that optimize for solution likelihood often yield incorrect solutions. To address this issue, we propose Guiding chain-of-thought ReAsoning with a CorrectnEss Discriminator (GRACE), a stepwise decoding approach that steers the decoding process towards producing correct reasoning steps. GRACE employs a step-level verifier or discriminator trained with a contrastive loss over correct and incorrect steps, which is used during decoding to score next-step candidates based on their correctness. Importantly, GRACE only requires sampling from the LM, without the need for LM training or fine-tuning. Using models from FLAN-T5 and LLaMA families, we evaluate GRACE over four math and two symbolic reasoning tasks, where it exhibits substantial performance gains compared to greedy decoding, verifiers, and self-consistency in most settings. When further combined with self-consistency, GRACE outperforms all the baselines by sizeable margins. Human and LLM evaluations over GSM8K show that GRACE not only improves the final answer accuracy but also the correctness of the intermediate reasoning. Our implementation can be accessed at https://github.com/mukhal/grace.