cs.LGOct 21, 2026

Parallelizable Probabilistic Latent Reasoning for Diverse Reasoning Trajectories

Authors: Seokhyun Lee, Seongjun Lee, Changhee Lee

Organizations: Korea University

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Language models relying on verbalized Chain-of-Thought (CoT) are constrained to token-by-token generation in the language space, where many tokens serve textual coherence rather than core reasoning. This entangles genuine differences in logical trajectories with superficial variations in verbalization. Moving reasoning into a compact, continuous latent space offers a promising alternative for efficiently representing and exploring diverse reasoning trajectories. However, current latent models remain largely deterministic, capturing only a single reasoning trajectory rather than modeling a distribution over multiple valid alternatives. While stochastic approaches parameterize distributions over latent reasoning steps, they frequently suffer from unstable optimization and limited empirical performance. Furthermore, their reliance on autoregressive latent prediction during training precludes effective parallelization, leading to prohibitively long training times. To overcome these bottlenecks, we propose Probabilistic Latent Reasoning Models (PLRM), a parallelizable variational framework for probabilistic latent reasoning. PLRM adopts a JEPA-style objective in which the posterior infers latent distributions in parallel from observed reasoning trajectories, while the prior learns to predict the subsequent latent distribution via teacher forcing. To model heterogeneous reasoning processes, we introduce a discrete strategy variable together with a mixture-based prior, allowing the model to preserve the Gaussian-mixture structure induced by multiple distinct reasoning trajectories rather than collapsing them into a single Gaussian. Across diverse benchmarks, PLRM achieves performance competitive with deterministic latent reasoning models on single-trajectory datasets while enabling efficient training and consistently outperforming competing stochastic baselines on multi-trajectory datasets.

Figures & tables

Table 1
Figure 1: Average training time per iteration as the number of reasoning steps increases. For CoT-SFT, each step corresponds to four tokens, reflecting its approximately 4×4\times longer reasoning sequences.
MethodsAcc. (%)
Fixed Posterior18.35
+ Answer Loss19.49
CoLaR21.62
PLRM (Ours)43.40
Table 3: Ablation study of PLRM.
Figure 2: PCA visualization of prior and posterior latent distributions under mono (M=1) and double-prior (M=2) settings.
MetricCoT-SFTReGuLaRCoLaRPLRM
Hit Rate0.91160.86650.73970.9205
Avg. TV Score0.61600.47530.36380.7097
Table 4: Hit rate and average TV Score across different methods.
Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 3: Overall Training Computation
 GPT-2Llama-3.2-1B
HyperparameterGSM8KGSM8K-NLCQAGSM8KGSM8K-NLCQA
Optimization
Learning rate1×10−31\times 10^{-3}1×10−31\times 10^{-3}2×10−32\times 10^{-3}3×10−43\times 10^{-4}3×10−43\times 10^{-4}3×10−43\times 10^{-4}
Epochs404040404010
Effective batch size128128128256256256
Latent Configuration
Appendix
Table 6: Main training and method-specific hyperparameters of PLRM. PTL denotes the number of posterior-visible CoT tokens per latent step.
Split# Unique Inputs# Valid Branches
Train21,46652,734
Validation3,1797,712
Test6,12815,101
Total30,77375,547
Appendix
Table 7: Statistics of the HiddenBranch dataset. Each unique input corresponds to one source question, while valid branches denote distinct reasoning–answer pairs associated with that input.
TypeMethodGPT-2Llama-3.2-1B
GSM8K-NLGSM8KCSQAGSM8K-NLGSM8KCSQA
VerbalizedExplicit CoT34.844.136.954.161.668.2
DeterministicCOCONUT–34.1––––
CODI35.343.744.049.755.674.0
SIM-CoT–42.6––56.1–
StochasticReGuLaR–––20.234.9–
Appendix
Table 8: Accuracy (%) of GPT-2 and Llama-3.2-1B-Instruct on in-distribution tasks.
Figure 4: PCA visualization of prior and posterior latent distributions for a HiddenBranch example with two valid reasoning branches. Top: unimodal prior. Bottom: two-component strategy-conditioned prior.
Figure 5: PCA visualization of prior and posterior latent distributions for a HiddenBranch example with three valid reasoning branches. Top: unimodal prior. Bottom: three-component strategy-conditioned prior.
Figure 6: PCA visualization of prior and posterior latent distributions for a HiddenBranch example with four valid reasoning branches. Top: unimodal prior. Bottom: four-component strategy-conditioned prior.
Figure 7: Qualitative HiddenBranch example with two valid reasoning branches generated from different strategy modes.
Figure 8: Qualitative HiddenBranch example with three valid reasoning branches generated from different strategy modes.
Figure 9: Qualitative HiddenBranch example with four valid reasoning branches generated from different strategy modes.
More in this topic

Continuous Latent Reasoning

Continuous latent reasoning studies how models can compute through continuous hidden states instead of spelling out every reasoning step.

See all 80 papers

Explore similar work

CardsList
Apr 9, 2026

SeLaR: Selective Latent Reasoning in Large Language Models

Chain-of-Thought (CoT) has become a cornerstone of reasoning in large language models, yet its effectiveness is constrained by the limited expressiveness of discrete token sampling. Recent latent reasoning approaches attempt to alleviate this limitation by replacing discrete tokens with soft embeddings (probability-weighted mixtures of token embeddings) or hidden states, but they commonly suffer from two issues: (1) global activation injects perturbations into high-confidence steps, impairing reasoning stability; and (2) soft embeddings quickly collapse toward the highest-probability token, limiting exploration of alternative trajectories. To address these challenges, we propose SeLaR (Selective Latent Reasoning), a lightweight and training-free framework. SeLaR introduces an entropy-gated mechanism that activates soft embeddings only at low-confidence steps, while preserving discrete decoding at high-confidence steps. Additionally, we propose an entropy-aware contrastive regularization that pushes soft embeddings away from the dominant (highest-probability) token's direction, encouraging sustained exploration of multiple latent reasoning paths. Experiments on five reasoning benchmarks demonstrate that SeLaR consistently outperforms standard CoT and state-of-the-art training-free methods.

Renyu Fu, Guibo Luo

Guangdong Provincial Key Laboratory of Ultra High Definition Immersive Media Technology Shenzhen Graduate School, Peking University

CoT ReasoningContinuous Latent ReasoningReasoning Consistency in Language Models
AskPDF
Jun 28, 2026

Invariant Reasoning Directions in Latent Trajectories of Language Models

Latent reasoning models perform multi-step inference directly in hidden-state space, yet the structure of these latent reasoning trajectories remains poorly understood. We show that contrastive refinement signals between stronger and weaker reasoning trajectories exhibit a highly concentrated low-rank structure, while unconstrained latent updates remain sensitive to paraphrases, checkpoint choice, and trajectory perturbations. These observations suggest that latent reasoning trajectories contain stable invariant directions mixed with unstable instance-specific variation. We introduce Trajectory-Invariant Latent Refinement (TILR), a training-free intervention framework for identifying and manipulating stable reasoning directions in latent space. TILR first learns a low-rank invariant subspace from contrastive trajectory differences across inputs, then constrains latent interventions to this subspace while suppressing poorly aligned updates through an adaptive alignment gate. Across six reasoning benchmarks, we find that a small number of latent directions explain most variation between strong and weak reasoning trajectories. Interventions on these directions causally improve reasoning consistency and reduce trajectory instability under paraphrases and perturbations. TILR improves answer consistency under paraphrase by ~10% and reduces latent trajectory variance by up to 50%50\% while preserving reasoning accuracy. These results support a geometric view of latent reasoning in which transferable reasoning behavior emerges from stable low-dimensional structure within hidden-state trajectories.

Arun Vignesh Malarkkan, Manan Roy Choudhury, Utkarsh Byahut +3

Arizona State University, Tempe, Arizona.

Reasoning Consistency in Language ModelsImplicit Reasoning in Language ModelsCausal Interventions in Language ModelsLanguage Model RobustnessRepresentation Geometry in Language ModelsLatent Reasoning
AskPDF
May 10, 2026

RuPLaR : Efficient Latent Compression of LLM Reasoning Chains with Rule-Based Priors From Multi-Step to One-Step

The Chain-of-Thought (CoT) paradigm, while enhancing the interpretability of Large Language Models (LLMs), is constrained by the inefficiencies and expressive limits of natural language. Latent Chain-of-Thought (latent CoT) reasoning, which operates in a continuous latent space, offers a promising alternative but faces challenges from structural complexities in existing multi-step or multi-model paradigms, such as error propagation and coordination overhead. In this paper, we introduce One-Model One-Step, a novel compression framework for Latent Reasoning with Rule-Based Priors(RuPLaR) to address this challenge. Our method trains an LLM to autonomously generate latent reasoning tokens in a single training stage, guided by rule-based prior probability distributions, thereby eliminating cascaded processes and inter-model dependencies. To ensure reasoning quality, we design a joint training objective that enforces answer consistency via cross-entropy, aligns soft token distributions with rule-based priors via KL divergence (the Soft Thinking constraint), and adds a problem-thought semantic alignment constraint in the representation space. Extensive experiments show that our compression framework not only improves accuracy by 11.03% over existing latent CoT methods on GSM8k-Aug but also achieves this with minimal token usage, underscoring its effectiveness and extensibility. Code: https://github.com/xiaocen-luo/RuPLaR.

Xiaocheng Luo, Kang Wang, Zaifu Zhan +2

School of Computer Science and Technology, Soochow University, Suzhou, 215006, Jiangsu, PR China · Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, 55455, MN, USA

LLM CompressionCoT CompressionLLM ReasoningLatent CoT Reasoning
AskPDF