Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Liu, Bingyang Kelvin, Chen, Ziyu Patrick, Woodruff, David P.
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2512.19171
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910002883592192
author Liu, Bingyang Kelvin
Chen, Ziyu Patrick
Woodruff, David P.
author_facet Liu, Bingyang Kelvin
Chen, Ziyu Patrick
Woodruff, David P.
contents Current autoregressive language models couple high-level reasoning and low-level token generation into a single sequential process, making the reasoning trajectory vulnerable to compounding expression errors. We propose JEPA-Reasoner, a novel architectural paradigm that decouples these tasks using a Joint-Embedding Predictive Architecture (JEPA) for pure latent-space reasoning and a separate Talker module for linguistic reconstruction. By isolating the reasoning engine from the discrete token-sampling process, our architecture enables: (1) Error Containment, where token-level failures cannot propagate into the latent reasoning chain; (2) Continuous Guidance, providing the generator with access to the entire lossless reasoning trajectory; and (3) Representation of Uncertainty, allowing the model to maintain multiple hypotheses via mixed latent vectors. Controlled experiments on synthetic and natural language tasks demonstrate that this decoupling enables a 0.9B model to achieve a 149.5\% improvement in 8-shot GSM8K accuracy over a coupled Transformer baseline trained on identical data. This work shifts the focus from scaling coupled models to investigating decoupled architectures as a more robust foundation for complex reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2512_19171
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle JEPA-Reasoner: Decoupling Latent Reasoning from Token Generation
Liu, Bingyang Kelvin
Chen, Ziyu Patrick
Woodruff, David P.
Computation and Language
Current autoregressive language models couple high-level reasoning and low-level token generation into a single sequential process, making the reasoning trajectory vulnerable to compounding expression errors. We propose JEPA-Reasoner, a novel architectural paradigm that decouples these tasks using a Joint-Embedding Predictive Architecture (JEPA) for pure latent-space reasoning and a separate Talker module for linguistic reconstruction. By isolating the reasoning engine from the discrete token-sampling process, our architecture enables: (1) Error Containment, where token-level failures cannot propagate into the latent reasoning chain; (2) Continuous Guidance, providing the generator with access to the entire lossless reasoning trajectory; and (3) Representation of Uncertainty, allowing the model to maintain multiple hypotheses via mixed latent vectors. Controlled experiments on synthetic and natural language tasks demonstrate that this decoupling enables a 0.9B model to achieve a 149.5\% improvement in 8-shot GSM8K accuracy over a coupled Transformer baseline trained on identical data. This work shifts the focus from scaling coupled models to investigating decoupled architectures as a more robust foundation for complex reasoning.
title JEPA-Reasoner: Decoupling Latent Reasoning from Token Generation
topic Computation and Language
url https://arxiv.org/abs/2512.19171