CORTEXA
← Browse
arxivcs.AIcs.CL2026-07-18

Training Continuous Chain of Thought Models: A Tale of Two Regimes

Varun Yerram, He He, Eunsol Choi

Continuous Chain-of-Thought methods replace verbose reasoning traces with a short sequence of dense latent representations. Earlier continuous CoT methods indirectly supervise the latent representations such that its final state match that of verbose reasoning traces, requiring autoregressive, slow generation during training. We introduce C-MTP, a simpler, faster direct supervision approach that models each latent as an average of the embeddings in the CoT traces to be compressed. Our approach outperforms a prior direct supervision method that approximates the distribution of compressed tokens, and performs competitively to slower indirect supervision approaches in existing evaluation setup with simplified CoT traces (less than 100 tokens). Lastly, we extend the evaluation of Continuous CoT methods to complex tasks with longer reasoning traces ($\ge$ few hundreds reasoning tokens). We find both direct and indirect supervision training methods perform poorly (roughly 65\% performance drop) in this setting, revealing the limitations of current continuous CoT methods. The code and checkpoints are released at https://github.com/Varun221/cmtp_research

View free PDFSource page

Related papers

arxivcs.CLcs.AIcs.LG2026-07-23

Token Budget Saturation and Mechanistic Early Detection of Reasoning Non-Convergence in Chain-of-Thought Models

Renuka Oladri, Niveda Jawahar, Abdirisak Mohamed

Chain-of-thought reasoning models such as DeepSeek-R1-Distill-Qwen-7B exhibit a bimodal convergence pattern: generations either terminate within a token budget (converged) or exhaust it without reaching a conclusion (non-converged). We characterize this phenomenon empirically, sh…

View free PDFSource page
arxivcs.LGcs.AIcs.CL2026-07-01

CausalMix: Data Mixture as Causal Inference for Language Model Training

Zinan Tang, Yukun Zhang, Shaomian Zheng, Zhuoshi Pan, Qizhi Pei, Dingnan Jin, et al.

In Large Language Model (LLM) training, data mixing plays a pivotal role in determining model performance. Recent methods optimize mixture weights via proxy models, but they rely on the assumption of static data distributions. As a result, when the underlying data pool shifts, th…

View free PDFSource page
arxivcs.CLcs.AI2026-07-16

Answer-Conditioned Chains of Thought Degrade Verifiable-Reasoning Distillation in Large Language Models

Jungseob Lee, Seungyoon Lee, Suhyune Son, Dongyub Jude Lee, Sungbin Han, Sugyeong Eo, et al.

A standard recipe for distilling the reasoning ability of large language models (LLMs) is to sample chains of thought from the model, keep those that reach the correct final answer, and fine-tune on the survivors. When sampling fails, a common fix shows the generator the gold ans…

View free PDFSource page
arxivcs.AIcs.CLcs.LG2026-07-08

Length Penalties Make Chain-of-Thought Less Monitorable

Bryce Little

Length-penalized reinforcement learning can shorten chain-of-thought reasoning while hiding an influence that drives the model's answer. In our experiments, training with length penalties does not stop misleading hints from steering models, even though the models' chains of thoug…

View free PDFSource page
arxivcs.AIcs.CLcs.LO2026-07-11

Interventional Grounding Audits: Black-Box Premise-Dependency Tests for LLM Chain-of-Thought via Predicate Substitution

Hironao Nakamura

Large language models produce chain-of-thought (CoT) reasoning that appears logically sound yet may not genuinely depend on its stated premises. We introduce interventional grounding audits, a black-box, step-level test of premise dependency: we intervene on a single premise by s…

View free PDFSource page