# Bagel-Robotics-1 inference adapter Bagel-Robotics-1 combines task-critical data selection, adaptive structured pruning and recovery training to produce a compact world-action model. VPP2 is used only as the base model for validation; the training and inference recipe is specific to Bagel-Robotics-1. This folder is the standalone XPolicyLab adapter for checkpoint 1000. It contains inference code, configuration and evaluation instructions. Model tensors download separately from [`bageldotcom/Bagel-Robotics-1`](https://huggingface.co/bageldotcom/Bagel-Robotics-1), pinned to `2651631c5d08819088475e8778676f5cf6d9afcf` and verified against `policy/Bagel_Robotics_1/artifact-manifest.json`. Install this adapter into an existing XPolicyLab checkout (the installer refuses to overwrite an existing policy): ```bash python install_adapter.py /absolute/path/to/XPolicyLab ``` Then follow [`policy/Bagel_Robotics_1/README.md`](policy/Bagel_Robotics_1/README.md) for the Python 3.10 environment, exact checkpoint and encoder download, and debug checks. [`REEVALUATION.md`](policy/Bagel_Robotics_1/REEVALUATION.md) gives native simulator task, suite and three-seed commands. RoboDojo and its assets are installed separately. The adapter is prepared against XPolicyLab `6baccd57ff958570d0db1588370289ee36c8bb70`; source and dependency pins are preserved in the policy folder. The training, pruning and export implementation is publicly available in the [Bagel-Robotics-1 repository](https://github.com/bageldotcom/Bagel-Robotics-1). The technical report remains planned for November 20, 2026. The repository is public and ungated. The pinned checkpoint and inference files can be downloaded without a Hugging Face account or access token.