LATIDIA · Investigación

Aprender de la experiencia de interacción a través de la destilación del contexto localizado

arXiv: 2610.21427v1Tipo de anuncio: nuevo Resumen: Aprender de la experiencia de interacción es esencial para permitir que los modelos de lenguaje grandes mejoren continuamente durante la implementación. La destilación contextual ofrece un camino prometedor:

La noticia

arXiv:2610.21427v1 Announce Type: new Abstract: Learning from interaction experience is essential for enabling large language models to continually improve during deployment. Context distillation offers a promising path: privileged information extracted from experience is provided to a teacher, whose behavior is then distilled into a student model that has no access to it. However, both existing forms of context distillation have limitations. Off-policy context distillation trains the student on teacher-generated trajectories and thus suffers from exposure bias; whereas on-policy context distillation trains on the student's own rollouts, whose states are often uninformative and fail to convey behaviors the student cannot yet produce. To resolve this dilemma, we introduce Localized Context Distillation