--- license: mit datasets: - lerobot/pusht_image tags: - lerobot - pusht - diffusion --- # Model Card for Mini Diffusion Policy / PushT We add few lines to add an extra level-2 minibatch for Diffusion Policy (as per [Mini Diffuser (ICRA 2026)](https://arxiv.org/abs/2505.09430)) trained for the `PushT` environment from [gym-pusht](https://github.com/huggingface/gym-pusht). This enables tens of equivalent batch size per gradient step, and end up saving at least 60% of the training time to obtain similar training results. ## How to Get Started with the Model See the [LeRobot library](https://github.com/huggingface/lerobot) for instructions on how to load and evaluate this model. ## Training Details Trained with a forked [LeRobot@ 7bd533a](https://github.com/utomm/lerobot/tree/minidp-0.4.2). The model was trained using [LeRobot's training script](https://github.com/huggingface/lerobot/blob/main/lerobot/scripts/train.py) and with the [pusht](https://huggingface.co/datasets/lerobot/pusht) dataset, using this command: ```bash lerobot-train --policy.type=minidiffusion --dataset.repo_id=lerobot/pusht_image\ --env.type=pusht --seed=100000 --batch_size=32 --log_freq=200 --wandb.disable_artifact=true\ --steps=100000 --eval_freq=10000 --save_freq=10000 --wandb.enable=true --policy.repo_id=id\ --wandb.project=minidp --policy.push_to_hub=false --policy.level2_batch_size=8 --job_name=minidp-32-8 ``` The training curves, and the comparasions with original DP may be found at https://api.wandb.ai/links/hu2240877635/defcr4wu