AI Navigate

トークンレベル適応潜在Chain-of-Thoughtによる事前学習

arXiv cs.CL / 2026/3/11

Ideas & Deep AnalysisModels & Research

要点

  • 本論文では、言語モデルの学習時に各トークンごとに内部推論の長さを動的に調整する、新しい事前学習手法「トークンレベル適応潜在Chain-of-Thought(adaptive latent CoT)」を提案する。
  • この手法は、可変長のCoT軌道を内在化することでモデルの効率性を向上させ、難しいトークンにはより長い推論経路を、容易なトークンには短い(あるいはゼロ長の)推論経路を割り当てる。
  • adaptive latent CoTは追加段階を必要とせず標準の事前学習から自然に出現し、トークン単位の適応停止により学習および推論の計算負荷を低減する。
  • Llamaモデルでの実験により、これまでの再帰的CoTベース手法よりも少ないFLOPsで、言語モデルの困惑度と下流タスクの精度が向上することが示された。
  • 本手法は、高品質データの不足や通信コスト増大という課題に対し、パラメータ数やデータ量を拡大する代わりにトークンごとの計算量を増やすことで対応している。

Computer Science > Computation and Language

arXiv:2602.08220 (cs)
[Submitted on 9 Feb 2026 (v1), last revised 10 Mar 2026 (this version, v2)]

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

View a PDF of the paper titled Pretraining with Token-Level Adaptive Latent Chain-of-Thought, by Boyi Zeng and 10 other authors
View PDF HTML (experimental)
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.
Comments:
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2602.08220 [cs.CL]
  (or arXiv:2602.08220v2 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2602.08220
Focus to learn more
arXiv-issued DOI via DataCite

Submission history

From: Yiqin Hao [view email]
[v1] Mon, 9 Feb 2026 02:49:15 UTC (5,197 KB)
[v2] Tue, 10 Mar 2026 05:18:24 UTC (5,197 KB)
Full-text links:

Access Paper:

Current browse context:
cs.CL
< prev   |   next >
Change to browse by:
cs

References & Citations

export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo
Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
Links to Code Toggle
Papers with Code (What is Papers with Code?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos

Demos

Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers

Recommenders and Search Tools

Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.