让大模型推理更高效，通过并行迭代减少计算时间。
Parallel Continuous Chain-of-Thought with Jacobi Iteration
- 用雅可比迭代并行更新思维潜变量，打破顺序依赖
- 训练和推理时间减少近50%，性能相当或更优
- 提升训练稳定性，适合追求效率的推理场景
连续思维链（Continuous CoT）通过连续潜变量实现紧凑的隐式推理，显著减少大语言模型的推理令牌消耗。然而，潜变量之间的序列依赖限制了并行训练，导致训练时间过长。本文提出并行连续思维链（PCCoT），对潜变量采用雅可比迭代方式，在每轮中并行更新所有变量，而非串行处理，从而大幅提升训练与推理效率。实验表明，合理设置迭代次数后，PCCoT在保持甚至超越原有性能的同时，可节省近50%的训练与推理时间。此外，该方法在训练过程中表现出更强的稳定性和鲁棒性。代码已开源：https://github.com/whyNLP/PCCoT。
原文摘要 · Abstract (English)
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.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。