Latent Thoughts Tuning: Bridging Context and Reasoning with Fused Information in Latent Tokens

TL;DR

Latent Thoughts Tuning (LT-Tuning) enhances latent reasoning in large language models by combining contextual and semantic information, improving robustness and accuracy over previous methods.

Contribution

The paper introduces LT-Tuning, a novel framework that fuses contextual and semantic cues for latent reasoning, addressing feature collapse and stability issues.

Findings

Outperforms existing latent reasoning methods.

Mitigates feature collapse and instability.

Achieves robust reasoning accuracy.

Abstract

While explicit Chain-of-Thought (CoT) equips Large Language Models (LLMs) with strong reasoning capabilities, it requires models to verbalize every intermediate step in text tokens, constraining the model thoughts to the discrete vocabulary space. Recently, reasoning in continuous latent space has emerged as a promising alternative, enabling more robust inference and flexible computation beyond discrete token constraints. However, current latent paradigms often suffer from feature collapse and instability, stemming from distribution mismatches when recurrently using hidden states as the input embeddings, or alignment issues when relying on assistant models. To address this, we propose Latent Thoughts Tuning (LT-Tuning), a framework that redefines how latent thoughts are constructed and deployed. Instead of relying solely on raw hidden states, our method introduces a Context-Prediction-Fusion mechanism that jointly leveraging contextual hidden states and predictive semantic guidance from the vocabulary embedding space. Combined with a progressive three-stage curriculum learning pipeline, LT-Tuning also enables dynamically switching between latent and explicit thinking modes. Experiments demonstrate that our method outperforms existing latent reasoning baselines, effectively mitigating feature collapse and achieving robust reasoning accuracy.