66
v1v2 (latest)

Pretraining with Token-Level Adaptive Latent Chain-of-Thought

Boyi Zeng
Yiqin Hao
He Li
Shixiang Song
Feichen Song
Zitong Wang
Siyuan Huang
Yi Xu
ZiWei He
Xinbing Wang
Zhouhan Lin
Main:12 Pages
8 Figures
Bibliography:3 Pages
2 Tables
Abstract

Scaling large language models by increasing parameters and training data is increasingly constrained by limited high-quality corpora and rising communication costs. This work explores an alternative axis: increasing per-token computation without expanding parameters, by internalizing latent Chain-of-Thought (CoT) into pretraining. We propose Pretraining with Token-Level Adaptive Latent CoT (adaptive latent CoT), where the model generates a variable-length latent CoT trajectory before emitting each token -- allocating longer trajectories to difficult tokens and shorter (or even zero) trajectories to easy ones. Importantly, this behavior emerges naturally from one-stage pretraining on general text and reduces computation in both training and inference via token-wise adaptive halting. Experiments with Llama architectures show that adaptive latent CoT consistently improves language modeling perplexity and broad downstream accuracy, even with fewer training FLOPs than prior recurrent baselines.

View on arXiv
Comments on this paper