Prompt Tuning Decision Transformers with Structured and Scalable Bandits

This is the official code repository for our NeurIPS 2025 paper "Prompt Tuning Decision Transformers with Structured and Scalable Bandits". Please cite this work as

@inproceedings{rietz2025neurips,
 author = {Rietz, Finn and Smirnov, Oleg and Karimi, Sara and Cao, Lele},
 booktitle = {Advances in Neural Information Processing Systems},
 title = {Prompt Tuning Decision Transformers with Structured and Scalable Bandits},
 volume = {38},
 year = {2025}
}

Instructions

Please see the instructions for installation and running each experiment in the corresponding directory:

• bandit_regret_experiment for the synthethic regret comparison of standard and our structured bandit for prompt tuning
• prompt_tune_2d_exeriment for the 2D experiments
• prompt_tune_mujoco_experiment for the MuJoCo experiments