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

• URL: http://arxiv.org/abs/2602.08220v1
• Date: Mon, 09 Feb 2026 02:49:15 GMT
• Title: Pretraining with Token-Level Adaptive Latent Chain-of-Thought
• Authors: Boyi Zeng, Yiqin Hao, He Li, Shixiang Song, Feichen Song, Zitong Wang, Siyuan Huang, Yi Xu, ZiWei He, Xinbing Wang, Zhouhan Lin,
• Abstract summary: Scaling large language models by increasing parameters and training data is increasingly constrained by limited high-quality corpora and rising communication costs.<n>This work explores an alternative axis: increasing per-token computation without expanding parameters, by internalizing latent Chain-of-Thought (CoT) into pretraining.<n>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.<n>Experiments with Llama architectures show that adaptive latent CoT consistently improves language modeling perplexity and broad downstream accuracy, even with fewer training FL
• Score: 44.19871205975474
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• 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.

Related papers

• AdaPonderLM: Gated Pondering Language Models with Token-Wise Adaptive Depth [23.442686851761298]
  AdaPonderLM is a self-supervised recurrent language model that learns token-wise early exiting during pretraining.<n>AdaPonderLM reduces inference compute at about 10% while maintaining comparable language modeling perplexity and competitive downstream accuracy.
  arXiv  Detail & Related papers  (2026-03-02T14:28:16Z)
• Entropy-Guided Token Dropout: Training Autoregressive Language Models with Limited Domain Data [89.96277093034547]
  We introduce EntroDrop, an entropy-guided token dropout method that functions as structured data regularization.<n>We show that EntroDrop consistently outperforms standard regularization baselines and maintains robust performance throughout extended multi-epoch training.
  arXiv  Detail & Related papers  (2025-12-29T12:35:51Z)
• Continuous Autoregressive Language Models [56.49239051750678]
  We introduce Continuous Autoregressive Language Models (CALM)<n>CALM uses a high-fidelity autoencoder to compress a chunk of K tokens into a single continuous vector.<n>We develop a comprehensive likelihood-free framework that enables robust training, evaluation, and controllable sampling.
  arXiv  Detail & Related papers  (2025-10-31T17:58:11Z)
• Test time training enhances in-context learning of nonlinear functions [51.56484100374058]
  Test-time training (TTT) enhances model performance by explicitly updating designated parameters prior to each prediction.<n>We investigate the combination of TTT with in-context learning (ICL), where the model is given a few examples from the target distribution at inference time.
  arXiv  Detail & Related papers  (2025-09-30T03:56:44Z)
• PonderLM-2: Pretraining LLM with Latent Thoughts in Continuous Space [44.24277388571869]
  We propose a novel pre-training methodology: Pretraining Language Models with Latent Thoughts (PonderLM-2)<n>Our approach pretrains a language model (LM) to first generate an intermediate latent thought-the last hidden state of the current position-which is then used as input to predict the actual subsequent token.<n>Experiments demonstrate that, at an identical inference cost, a LM that generates one additional latent thought per token outperforms a standard model with double the parameters.
  arXiv  Detail & Related papers  (2025-09-27T08:38:08Z)
• READER: Retrieval-Assisted Drafter for Efficient LLM Inference [0.0386965802948046]
  Autoregressive Language Models instantiate a factorized likelihood over token sequences, yet their strictly sequential decoding process imposes an intrinsic lower bound on latency inference.<n>This bottleneck has emerged as a central obstacle to the scalable deployment of large-scale generative models.<n>We present READER, a speculative decoding framework that bypasses the training of the auxiliary draft model.
  arXiv  Detail & Related papers  (2025-08-12T16:47:48Z)
• TAH-QUANT: Effective Activation Quantization in Pipeline Parallelism over Slow Network [21.231881562816373]
  We introduce TAH-Quant (Tile-wise Adaptive Hadamard Quantization), a novel activation quantization framework designed specifically for pipeline parallelism.<n>Our approach integrates fine-grained tile-wise quantization for precise control, entropy-guided token-level adaptive bit allocation for optimal bit usage, and a Hadamard-based transform with pivot element swapping to effectively suppress quantization outliers.
  arXiv  Detail & Related papers  (2025-06-02T06:13:41Z)
• LESA: Learnable LLM Layer Scaling-Up [57.0510934286449]
  Training Large Language Models (LLMs) from scratch requires immense computational resources, making it prohibitively expensive.<n>Model scaling-up offers a promising solution by leveraging the parameters of smaller models to create larger ones.<n>We propose textbfLESA, a novel learnable method for depth scaling-up.
  arXiv  Detail & Related papers  (2025-02-19T14:58:48Z)
• Enhancing Robustness of Vision-Language Models through Orthogonality Learning and Self-Regularization [77.62516752323207]
  We introduce an orthogonal fine-tuning method for efficiently fine-tuning pretrained weights and enabling enhanced robustness and generalization. A self-regularization strategy is further exploited to maintain the stability in terms of zero-shot generalization of VLMs, dubbed OrthSR. For the first time, we revisit the CLIP and CoOp with our method to effectively improve the model on few-shot image classficiation scenario.
  arXiv  Detail & Related papers  (2024-07-11T10:35:53Z)
• Towards Continual Learning Desiderata via HSIC-Bottleneck Orthogonalization and Equiangular Embedding [55.107555305760954]
  We propose a conceptually simple yet effective method that attributes forgetting to layer-wise parameter overwriting and the resulting decision boundary distortion. Our method achieves competitive accuracy performance, even with absolute superiority of zero exemplar buffer and 1.02x the base model.
  arXiv  Detail & Related papers  (2024-01-17T09:01:29Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.