paper-with-me

Papers

CODI: Compressing Chain-of-Thought into Continuous Space via Self-Distillation

2025-02-28 · Zhenyi Shen, Hanqi Yan, Linhai Zhang, Zhanghao Hu, Yali Du, Yulan He

Chain-of-Thought (CoT) reasoning enhances Large Language Models (LLMs) by encouraging step-by-step reasoning in natural language. However, leveraging a latent continuous space for reasoning may offer benefits in terms of both efficiency and robustness. Prior implicit CoT methods attempt to bypass language completely by reasoning in continuous space but have consistently underperformed compared to the standard explicit CoT approach. We introduce CODI (Continuous Chain-of-Thought via Self-Distillation), a novel training framework that effectively compresses natural language CoT into continuous space. CODI jointly trains a teacher task (Explicit CoT) and a student task (Implicit CoT), distilling the reasoning ability from language into continuous space by aligning the hidden states of a designated token. Our experiments show that CODI is the first implicit CoT approach to match the performance of explicit CoT on GSM8k at the GPT-2 scale, achieving a 3.1x compression rate and outperforming the previous state-of-the-art by 28.2% in accuracy. CODI also demonstrates robustness, generalizable to complex datasets, and interpretability. These results validate that LLMs can reason effectively not only in natural language, but also in a latent continuous space.

📄 PDF Abstract BibTeX arXiv:2502.21074

Code (1)

guanziju/Open_CODI pytorch

Tasks

GSM8K

Methods 이 논문이 사용한 방법론

Attention 설명 없음
Cosine Annealing Cosine Annealing is a type of learning rate schedule that has the effect of starting with a large learning rate that is relatively rapidly decreased to a minimum value before…
Linear Warmup With Cosine Annealing Linear Warmup With Cosine Annealing is a learning rate schedule where we increase the learning rate linearly for $n$ updates and then anneal according to a cosine schedule…
Softmax The Softmax output function transforms a previous layer's output into a vector of probabilities. It is commonly used for multiclass classification. Given an input vector $x$…
Attention Dropout Attention Dropout is a type of dropout used in attention-based architectures, where elements are randomly dropped out of the…
Refunds@Expedia|||How do I get a full refund from Expedia? “How do I get a full refund from Expedia? How do I get a full refund from Expedia? – Call ☎️ +1-(888) 829 (0881) or +1-805-330-4056 or +1-805-330-4056 for Quick Help &…
Linear Layer A Linear Layer is a projection $\mathbf{XW + b}$.
Residual Connection 설명 없음

Similar Papers 제목 키워드 기반

Is continuous CoT better suited for multi-lingual reasoning?

2026-03-09 · Ali Hamza Bashir, Behzad Shomali, Markus Frey, Mehdi Ali 외 arxiv

We investigate whether performing reasoning in a continuous latent space leads to more robust multilingual capabilities. We compare Continuous Chain-of-Thought (using the CODI framework) against standard supervised fine-…

ImgCoT: Compressing Long Chain of Thought into Compact Visual Tokens for Efficient Reasoning of Large Language Model

2026-01-30 · Xiaoshu Chen, Sihang Zhou, Ke Liang, Taichun Zhou 외 arxiv

Compressing long chains of thought (CoT) into compact latent tokens is crucial for efficient reasoning with large language models (LLMs). Recent studies employ autoencoders to achieve this by reconstructing textual CoT f…

PPC-GPT: Federated Task-Specific Compression of Large Language Models via Pruning and Chain-of-Thought Distillation

2025-02-21 · Tao Fan, Guoqiang Ma, Yuanfeng Song, Lixin Fan 외

Compressing Large Language Models (LLMs) into task-specific Small Language Models (SLMs) encounters two significant challenges: safeguarding domain-specific knowledge privacy and managing limited resources. To tackle the…

Knowledge DistillationPrivacy PreservingText Generation

SoftCoT: Soft Chain-of-Thought for Efficient Reasoning with LLMs

2025-02-17 · Yige Xu, Xu Guo, Zhiwei Zeng, Chunyan Miao

Chain-of-Thought (CoT) reasoning enables Large Language Models (LLMs) to solve complex reasoning tasks by generating intermediate reasoning steps. However, most existing approaches focus on hard token decoding, which con…

parameter-efficient fine-tuning

Knowledge Distillation for Large Language Models

2026-03-14 · Alejandro Paredes La Torre, Barbara Flores, Diego Rodriguez arxiv

We propose a resource-efficient framework for compressing large language models through knowledge distillation, combined with guided chain-of-thought reinforcement learning. Using Qwen 3B as the teacher and Qwen 0.5B as …

Knowledge DistillationReinforcement Learning