LaT: LLM-as-Trainer for Multi-Task Vehicle Routing Solvers

2026-07-20

AI Summary

This work addresses the challenge that multi-task vehicle routing problem (VRP) solvers, when trained jointly, are hindered by disparities in optimization difficulty across variants and the absence of dynamic feedback mechanisms. To overcome this, the paper introduces the LLM-as-Trainer (LaT) paradigm, which, for the first time, leverages a pretrained large language model as an external trainer in combinatorial optimization. The approach periodically analyzes cross-task validation metrics to generate phased guidance vectors, which are fused with constraint vectors and injected into the encoder to steer policy optimization—enabling adaptive multi-task training without additional gradient updates. Evaluated on 16 VRP variants, the method significantly enhances solver performance and demonstrates strong generalization on both seen and unseen problem instances.

Abstract

Multi-task neural solvers aim to handle multiple Vehicle Routing Problem (VRP) variants within a unified model, avoiding separate training for each constraint combination. However, VRP variants differ in optimization difficulty, while existing methods lack stage-wise feedback on their training status, making the model biased to some specific variants. Although meta-learning can support adaptive training, it typically requires bi-level optimization and additional gradient updates, increasing computational cost. To address this limitation, we propose LLM-as-Trainer (LaT), a plug-and-play training paradigm that uses a pretrained large language model as an external trainer. LaT periodically analyzes cross-task validation metrics to generate a stage-wise guidance vector. This vector is combined with the current task's constraint vector and injected into each encoder layer, providing the neural solver with additional training information during subsequent policy optimization. Experiments on 16 VRP variants show that LaT improves the solution quality of several state-of-the-art multi-task neural solvers on both trained and unseen variants, supporting the effectiveness and generality of the proposed training paradigm.

Problem

• Vehicle Routing Problem
• multi-task learning
• training bias
• optimization difficulty
• neural solvers

Innovation

• LLM-as-Trainer
• multi-task learning
• vehicle routing problem
• adaptive training
• neural combinatorial optimization