用平均嵌入直接训练连续思维链，速度更快且效果不差。
Training Continuous Chain of Thought Models: A Tale of Two Regimes
- 用推理痕迹嵌入的平均值直接监督隐状态，避免自回归生成。
- 在少于100词的简化推理中表现接近慢速间接方法。
- 在数百词长推理任务中性能骤降65%，暴露现有方法局限。
连续思维链（Continuous Chain-of-Thought）方法将冗长的推理过程替换为短序列的稠密隐状态。早期方法通过间接监督使隐状态最终匹配原始推理痕迹，需自回归、缓慢训练。本文提出C-MTP，一种更简单快速的直接监督方法：将每个隐状态建模为待压缩推理痕迹嵌入的平均值。该方法优于先前的直接监督方法，且在简化推理（少于100个词）的现有评估中表现与更慢的间接方法相当。进一步扩展至含数百词复杂推理的任务，发现直接与间接监督方法均性能下降约65%，揭示当前连续思维链方法的瓶颈。代码与模型权重已公开于https://github.com/Varun221/cmtp_research。
原文摘要 · Abstract (English)
Continuous Chain-of-Thought methods replace verbose reasoning traces with a short sequence of dense latent representations. Earlier continuous CoT methods indirectly supervise the latent representations such that its final state match that of verbose reasoning traces, requiring autoregressive, slow generation during training. We introduce C-MTP, a simpler, faster direct supervision approach that models each latent as an average of the embeddings in the CoT traces to be compressed. Our approach outperforms a prior direct supervision method that approximates the distribution of compressed tokens, and performs competitively to slower indirect supervision approaches in existing evaluation setup with simplified CoT traces (less than 100 tokens). Lastly, we extend the evaluation of Continuous CoT methods to complex tasks with longer reasoning traces ($\ge$ few hundreds reasoning tokens). We find both direct and indirect supervision training methods perform poorly (roughly 65\% performance drop) in this setting, revealing the limitations of current continuous CoT methods. The code and checkpoints are released at https://github.com/Varun221/cmtp_research
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。