Search papers, labs, and topics across Lattice.
The paper introduces Latent Thoughts Tuning (LT-Tuning), a novel framework for latent space reasoning in LLMs that addresses feature collapse and instability issues in existing methods. LT-Tuning leverages a Context-Prediction-Fusion mechanism that combines contextual hidden states with predictive semantic guidance from the vocabulary embedding space to construct more robust latent thoughts. Experiments show that LT-Tuning outperforms existing latent reasoning baselines by dynamically switching between latent and explicit thinking modes using a progressive three-stage curriculum learning pipeline.
LLMs can reason more robustly by fusing contextual hidden states with vocabulary embedding guidance, enabling dynamic switching between latent and explicit thinking.
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.