Parallel Continuous Chain-of-Thought with Jacobi Iteration

TL;DR

This paper introduces PCCoT, a parallel training method for continuous chain-of-thought reasoning in large language models, significantly reducing training time while maintaining or improving performance.

Contribution

It proposes Jacobi iteration for parallel updating of latent thought tokens, enabling efficient training and inference in continuous CoT models.

Findings

Achieves nearly 50% reduction in training and inference time.

Maintains or improves reasoning performance with proper iteration count.

Demonstrates enhanced stability and robustness during training.

Abstract

Continuous chain-of-thought has been shown to be effective in saving reasoning tokens for large language models. By reasoning with continuous latent thought tokens, continuous CoT is able to perform implicit reasoning in a compact manner. However, the sequential dependencies between latent thought tokens spoil parallel training, leading to long training time. In this paper, we propose Parallel Continuous Chain-of-Thought (PCCoT), which performs Jacobi iteration on the latent thought tokens, updating them iteratively in parallel instead of sequentially and thus improving both training and inference efficiency of continuous CoT. Experiments demonstrate that by choosing the proper number of iterations, we are able to achieve comparable or even better performance while saving nearly 50% of the training and inference time. Moreover, PCCoT shows better stability and robustness in the training process. Our code is available at https://github.com/whyNLP/PCCoT.