Compressed Chain of Thought: Efficient Reasoning Through Dense Representations
This addresses efficiency and performance issues in reasoning tasks for language model users, though it is incremental as it builds on existing contemplation token methods.
The paper tackles the high latency of chain-of-thought decoding in language models by proposing Compressed Chain-of-Thought (CCoT), a framework that generates variable-length, continuous contemplation tokens as compressed reasoning representations, achieving improvements in accuracy with adaptive control over token count.
Chain-of-thought (CoT) decoding enables language models to improve reasoning performance at the cost of high generation latency in decoding. Recent proposals have explored variants of contemplation tokens, a term we introduce that refers to special tokens used during inference to allow for extra computation. Prior work has considered fixed-length sequences drawn from a discrete set of embeddings as contemplation tokens. Here we propose Compressed Chain-of-Thought (CCoT), a framework to generate contentful and continuous contemplation tokens of variable sequence length. The generated contemplation tokens are compressed representations of explicit reasoning chains, and our method can be applied to off-the-shelf decoder language models. Through experiments, we illustrate how CCoT enables additional reasoning over dense contentful representations to achieve corresponding improvements in accuracy. Moreover, the reasoning improvements can be adaptively modified on demand by controlling the number of contemplation tokens generated.