Sub-goal Distillation: A Method to Improve Small Language Agents

arXiv:2405.02749v1 Announce Type: new
Abstract: While Large Language Models (LLMs) have demonstrated significant promise as agents in interactive tasks, their substantial computational requirements and restricted number of calls constrain their practical utility, especially in long-horizon interactive tasks such as decision-making or in scenarios involving continuous ongoing tasks. To address these constraints, we propose a method for transferring the performance of an LLM with billions of parameters to a much smaller language model (770M parameters). Our approach involves constructing a hierarchical …

abstract agents arxiv computational constraints continuous cs.lg decision distillation horizon interactive language language models large language large language models llms making practical requirements small tasks type utility while